1. Stock assessment for eastern Bering Sea walleye pollock
| This report may be cited as: Ianelli, J. et al. 2024. Assessment of the |
| eastern Bering Sea walleye pollock. North Pacific Fishery Management Council, |
| Anchorage, AK. Available here. |
Executive summary
This chapter covers the Eastern Bering Sea (EBS) region—the Aleutian Islands region (Chapter 1A) and the Bogoslof Island area (Chapter 1B) are presented separately.
A multi-species stock assessment is provided separately and available here. A list of this document contents, including tables and figures is provided in Section 17.
Summary of changes in assessment inputs
Relative to last year’s BSAI SAFE report, the following substantive changes have been made in the EBS pollock stock assessment. This includes the 2024 NMFS bottom-trawl survey (BTS) covering the EBS and NBS. As before, these data were treated with a spatio temporal model for index standardization. Age data from this survey effort was compiled and included (also with an extensive spatio-temporal model treatment). The NMFS acoustic-trawl survey (ATS) age composition data was revised from the preliminary estimates developed in 2022. The BTS chartered boats also collected acoustic data and the series was updated this year (AVO). Explorations were presented in J. Ianelli (2023).
Changes in the data
Observer data for catch-at-age and average weight-at-age from the 2023 fishery were finalized and included.
Total catch as reported by NMFS Alaska Regional office was updated and included through 2024.
In summer 2024, the AFSC conducted the bottom trawl survey in the EBS and extended into the NBS. A VAST model evaluation (including the cold-pool extent) was used as the main index.
We updated estimates of weight-at-age data used to compute spawning biomass as presented to the Plan Team and SSC in September/October 2023 (see J. Ianelli (2023) for details) including estimates to 2024.
We added a 2024 estimate to the time series from the acoustic data collected from the bottom trawl survey covering 2006-2024 (except for 2020). This represents the so-called “AVO” data (as presented in J. Ianelli (2023)).
We added a 2024 estimated biomass and preliminary age-composition from the 2024 ATS survey. The age-composition estimate was based on the BTS age-length key data plus a juvenile sample from the ATS.
Changes in the assessment methods
The assessment method was the same as presented in December of 2023 (J. Ianelli (2023)).
Summary of EBS pollock results
The results from the xxx 2022 assessment have largely been confirmed: the 2018 year class appears to be one of the most abundant on record. Nonetheless, the bottom-trawl survey was lower than expected (about 28% below the long-term mean and the tenth lowest over the 41-year survey period). The new AVO index (presented in September 2023) expanded the area covered by acoustics and provided more precision (lower CV in the point estimates) than in the past. Ancilliary data indicate that the pollock in 2023 are substantially skinnier than average given their length. The average weight-at-age was about average for the 2018 year class, but lighter for most other ages.
The following table is based on results from the selected model (“Model 23.0”) based on changes presented in J. Ianelli (2023). The ABC recommendation is based on Tier 3 calculation as a proxy for Tier 1 because of the variability indicated by the very high value based on the \(F_{MSY}\) estimate and the large but uncertain 2018 year class.
Response to SSC and Plan Team comments
SSC General groundfish stock assessment comments
The following are relevant SSC comments from their December 2022 minutes.
The SSC recommends that for future Tier 1-3 assessments some consideration be given as to how best to represent biomass estimates in the Executive Summary table for each stock (currently, model total biomass and spawning stock biomass are provided) so that the relationship of the biomass to the OFL and ABC in the stock status table is clear. - We agree. Within the document we include biomass estimates that are outcomes for ABC and OFL calculations. However, the estimates involve an application of expected age-specific selectivity which can be variable.
For all assessments using VAST, the SSC requests a figure comparing the VAST estimate used in the previous assessment to the current assessment (if new data are added), noting that VAST will refit the time series when additional data are added and the estimated extent and directionality of spatial correlation may change. - We include model comparisons showing the impact of new (updated) VAST time series.
The SSC suggests that walleye pollock is a good candidate for considering the impacts of highly variable recruitment on reference points in the context of the Council’s harvest control rules (see discussion on working groups in the JGPT report section). For example, the SAFE authors suggested exploring an explicit harvest control rule that maintains productivity at the level observed over recent decades (p. 33). The SSC supports considerations of modified harvest control rules, particularly for stocks with highly variable and uncertain recruitment. If the Council chooses, this could include considerations for stabilizing catches over time or including other economic considerations in the harvest control rules.
- For the 2023 assessment we examine the variability of the biological reference points historically and note that there is general stability in the \(B_{MSY}\) estimates. The Tier 1 ABC/OFL calculations can result in highly variable estimates as the stock approaches \(B_{MSY}\) and drops below that value (as happened in the 2009-2010 period).
The SSC had the following additional recommendations for the authors:
Maturity and growth information from the NBS has not been examined yet. Given the possible importance of the NBS to walleye pollock and other species in the future, the SSC suggests this should be a high priority.
- These data are being processed and this work is underway
The SSC supports efforts to implement recent advances in improving the statistical treatment of compositional data using the Dirichlet distribution or other approaches.
- Tradeoffs in data weighting were pursued in September 2023 and sought to find a balance between observation error and process errors. In general, trade-offs in data weighting appear consistent with sampling levels and include objective approaches to their specifications.
The SAFE document lists a number of research recommendations (p. 36/37). The SSC notes that some of these are at least in progress. The SSC generally supports these recommendations but requests that the authors update the list of priorities to clarify to what extent some of these priorities have been partially or fully addressed.
- We updated the priorities and listed those that have been completed or are continued to be underway
In particular, the SSC notes that genetic sample collection and analyses are listed as a research priority across all pollock stocks and that some work has been completed. The SSC highlights the importance of additional genetics work and would appreciate an update on the status of this work either as part of the assessment or separately.
- We revisited the stock structure work attached as an appendix to the 2015 SAFE report chapter and are examining the extent that this work needs updating. Updated genetics work indicate that the Bering Sea pollock represent a distinct stock. This work indicates that GOA pollock seems to be similar to some Aleutian pollock but some AI pollock are distinct (Spies and Schaal, pers. comm.).
The SSC appreciated the adjustments to weight-at-age in the survey that was included in this year’s assessment and suggests that these changes may be substantial enough to warrant an examination of their impact on assessment results.
- This publication has been completed and in the present assessment we evaluated the implication of alternative spawning biomass-at-age assumptions.
With respect to the multi-species CEATTLE model, the SSC concurs with Plan Team recommendations to use the model to inform risk table discussions and to consider ways in which model outputs, in particular estimates of predation mortality, can inform single-species assessments.
- We included some comments to this effect.
The SSC encourages the authors to consider model-based solutions to uncertain recruitment estimates rather than ad-hoc adjustments. In particular, reductions in the assumed recruitment variance parameter may result in less extreme recruitment estimates. Other systematic approaches to addressing the uncertainty may also be considered.
- We revisited applying the age-determination error matrix as a sensitivity as this can impact the recruitment variability and estimation uncertainty.
The SSC suggests that authors include a plot to compare estimates of recent recruitments as they change over time similar to Fig 3.33 (pg. 88) in the sablefish assessment.
- We provide a figure of estimated recruitment by year class (1977 – 2019) in number of age-1 fish (billions of fish) for the 2022 and 2023 models.
The SSC supports the move across assessment from design-based estimates of survey biomass to VAST estimates. The SSC recommends that the design-based estimates be produced as a check on VAST estimates and as a fallback option if needed, although they may not need to be included in the assessment.
- We provide a table showing the design-based estimates and conduct a model run with those estimates. This may be an approach to adopt so that bridging across assessment modeling platforms can be facilitated (most other assessment model platforms are unable to deal with index time series that have a covariance matrix)
The SSC noted that a consideration of whether the observed sensitivity in the SRR to prior specification should constitute an increased risk level specification within the assessment or population dynamics related considerations should be considered. This could provide a clearer justification for the use of the Tier 3 calculation as the basis for harvest specification.
- We evaluated factors affecting the Tier classification in the 2020 assessment and showed that the priors used reflect the SRR curve were conservative and justified based on residual patterns near the origin (as opposed to alternatives that fit data on the descending slope of the Ricker SRR.
The SSC recommends that if the assessment is considered in the appropriate Tier, buffers should be based on the use of the Risk Table rather than the continued use of Tier 3 calculations for a Tier 1 stock.
- We agree.
The SSC also notes that an alternative approach to consider for a buffer below the maximum permissible would be apply Tier 2 control rule. This tier uses the SR relationship for stock status and OFL, but uses the ratio of SPR rates for adjustments when the stock is below \(B_{MSY}\).
- An examination of Tier 2 as an option resulted in a value of 2,628,000 t (or a hybrid of Tier 1 and 2 of 2,259,000 t) for 2025 ABC values. We note that selecting Tier 2 would require similar reliance on the underlying productivity estimates (via the stock-recruitment relationship) and how that affects the reference fishing rate (\(F_{MSY}\)).
The SSC had a number of recommendations for additional research supporting this assessment:
From previous requests:
The SSC also looks forward to estimates of movement and abundance along the US-Russia EEZ boundary based on echosounders fixed to moorings in this area.
- The data evaluation from the moored sounders has been completed and initial results show that the flux of pollock back and forth over the maritime boundary is considerable, and appears to be a function of temperature conditions.
Introduction
General
Walleye pollock (Gadus chalcogrammus; hereafter referred to as pollock) are broadly distributed throughout the North Pacific with the largest concentrations found in the Eastern Bering Sea. Also known as Alaska pollock, this species continues to play important roles ecologically and economically.
Review of Life History
In the EBS pollock generally spawn during March-May and in relatively localized regions during specific periods (Bailey (2000) ). Generally spawning begins nearshore north of Unimak Island in March and April and later near the Pribilof Islands (Bacheler et al. (2010)). Females are batch spawners with up to 10 batches of eggs per female per year (during the peak spawning period). Eggs and larvae of EBS pollock are planktonic for a period of about 90 days and appear to be sensitive to environmental conditions. These conditions likely affect their dispersal into favorable areas (for subsequent separation from predators) and also affect general food requirements for over-wintering survival (Gann et al. (2015), Heintz et al. (2013), Hunt Jr. et al. (2011), Ciannelli, Brodeur, and Napp (2004)). Duffy-Anderson et al. (2016) provide a review of the early life history of EBS pollock.
Throughout their range juvenile pollock feed on a variety of planktonic crustaceans, including calanoid copepods and euphausiids. In the EBS shelf region, one-year-old pollock are found throughout the water column, but also commonly occur in the NMFS bottom trawl survey. Ages 2 and 3 year old pollock are rarely caught in summer bottom trawl survey gear and are more common in the midwater zone as detected by mid-water acoustic trawl surveys. Younger pollock are generally found in the more northern parts of the survey area and appear to move to the southeast as they age (Buckley, Greig, and Boldt (2009)). Euphausiids, principally Thysanoessa inermis and T. raschii, are among the most important prey items for pollock in the Bering Sea (Livingston (1991); Lang et al. (2000); Brodeur et al. (2002); Ciannelli, Brodeur, and Napp (2004); Lang, Livingston, and Dodd (2005)). Pollock diets become more piscivorous with age, and cannibalism has been commonly observed in this region. However, Buckley et al. (2015) showed spatial patterns of pollock foraging varies by size of predators. For example, the northern part of the shelf region between the 100 and 200 m isobaths (closest to the shelf break) tends to be more piscivorous than pollock found in more near-shore shallow areas.
Stock structure
Stock structure for EBS pollock was evaluated in James Ianelli, Kotwicki, and Honkalehto (2015). In that review past work on genetics (e.g., Bailey et al. (1999), Canino et al. (2005)) provided insight on genetic differentiation. The investigation also compared synchrony in year-classes and growth patterns by region. Pollock samples from areas including Zhemchug Canyon, Japan, Prince William Sound, Bogoslof, Shelikof, and the Northern Bering Sea were processed and results presented in J. Ianelli et al. (2021).
A group of researchers at AFSC led by Drs. Ingrid Spies and Sara Schaal have updated the recent genetics study and this is summarized here:
Adult samples of walleye pollock were collected from 15 locations spanning Japan, the Bering Sea, and the Gulf of Alaska and were used for genetic analysis. Researchers performed low-coverage whole genome sequencing on 547 individuals from these sampling locations, which resulted in roughly 2 million polymorphic loci found throughout the genome. Although genetic differentiation is subtle in walleye pollock compared to other species, there are two strong and notable genetic breaks that highlight the genetic stock structure present. The first is between all the US samples and Japan (Figure 1: the split on PC axis 1) and the second is between the Bering Sea and the GOA/Aleutian Islands (Figure 1: the split on PC axis 2). This suggests that Bering Sea walleye pollock are genetically distinct from GOA/Aleutian Islands walleye pollock. Walleye pollock in the Aleutian Islands and Bogoslof show weak divergence with the GOA. Individuals caught in Adak and Atka comprise two genetic groups (Figure 1: dark green points). One group is not differentiated from the rest of the GOA (Figure 1 : dark green points clustering behind the GOA points in pink) and the other shows some divergence along PC axis 1. This complex group warrants further investigation to understand the genomic regions that may be divergent between these two genetic groups present in the Adak and Atka. Additionally, individuals from Bogoslof show weak divergence from GOA samples (Figure 1: light green points). Walleye Pollock are currently managed as four distinct groups: 1) GOA, 2) Bering Sea, 3) Aleutian Islands and 4) Bogoslof. Our genetic groups mostly align with these delineations. However, the Aleutian Islands/Bogoslof stocks show subtle differentiation from the rest of the GOA.
For management purposes, the preliminary conclusions from these genetics results are: 1) there is stock structure in pollock that appears to be stable through time and 2) Some aspect of stock structure is latitudinal—Bering Sea pollock appear distinct from fish collected from the Gulf of Alaska and the Aleutian Islands. The results appear strong enough that a GTseq panel could be designed in the future to determine stock of origin of pollock, the scale of which may be relatively large, such as “Bering Sea” or “GOA”.
Fishery
Description of the directed fishery
Historically, EBS pollock catches were low until directed foreign fisheries began in 1964. Catches increased rapidly during the late 1960s and reached a peak in 1970–75 when they ranged from 1.3 to 1.9 million t annually. Following the peak catch in 1972, bilateral agreements with Japan and the USSR resulted in reductions. During a 10-year period, catches by foreign vessels operating in the “Donut Hole” region of the Aleutian Basin were substantial totaling nearly 7 million t (). A fishing moratorium for this area was enacted in 1993 and only trace amounts of pollock have been harvested from the Aleutian Basin region since then. Since the late 1970s, the average EBS pollock catch has been about 1.2 million t, ranging from 0.810 million t in 2009 to nearly 1.5 million t during 2002–2006 (). United States vessels began fishing for pollock in 1980 and by 1988 the fishery became fully domestic. The current observer program for the domestic fishery formally began in 1991 and prior to that, observers were deployed aboard the foreign and joint-venture operations since the late 1970s. From the period 1991 to 2011 about 80% of the catch was observed at sea or during dockside offloading. Since 2011, regulations require that all vessels participating in the pollock fishery carry at least one observer so nearly 100% of the pollock fishing operations are monitored by scientifically trained observers. Historical catch estimates used in the assessment, along with management measures (i.e., OFLs, ABCs and TACs) are shown in ().
Catch patterns
The “A-season” for directed EBS pollock fishing opens on January 20th and fishing typically extends into early-mid April. During this season the fishery targets pre-spawning pollock and produces pollock roe that, under optimal conditions, can comprise over 4% of the catch in weight. The summer, or “B-season” presently opens on June 10th and fishing extends through noon on November 1st. The A-season fishery concentrates primarily north and west of Unimak Island depending on ice conditions and fish distribution. There has also been effort along the 100m depth contour (and deeper) between Unimak Island and the Pribilof Islands. The general pattern by season (and area) has varied over time with recent B-season catches occurring in the southeast portion of the shelf (east of 170\(^\circ\)W longitude; Figure 2).
Since 2011, regulations and industry-based measures to reduce Chinook salmon bycatch have affected the spatial distribution of the fishery and to some degree, the way individual vessel operators fish (Stram and Ianelli (2014)). Comparing encounters of bycatch relative to the effort (total duration of all tows) the pollock fleet had a slight increase in the Chinook salmon bycatch rate (Figure 3). The nominal catch rate of sablefish in the pollock fishery continue to be above historical averages (Figure 3) while for herring, the rate was low compared to 2020.
The catch estimates by sex for the seasons indicate that over time, the number of males and females has been fairly equal but in the period 2017-2022 the A-season catch of females has been slightly higher and conversely, in the B-season there has been a slightly higher number of males taken (Figure 4). The pattern of catch numbers is impacted by the magnitude of the quota (e.g., the drop in 2022 when the TAC was lower) but also in the relative size of fish. For example, in 2020 estimated absolute numbers of catch were relatively high because fish were smaller (and younger) than average.
The 2023 A-season fishery spatial pattern had a relatively more catch around the Pribilof Islands compared to 2021 (Figure 5). The amount of fishing near the Pribilof Islands was lower than commonly observed in 2022. The 2023 A-season nominal catch rates were near peak levels for all fleet sectors (middle panel, Figure 6). Beginning in 2017, due to a regulatory change, up to 45% of the TAC could be taken in the A-season (previously only 40% of the TAC could be taken). This conservation measure was made to allow greater flexibility to avoid Chinook salmon in the B-season. The pollock fleet as a whole continues to take advantage of this flexibility (Figure 7). This figure shows that the proportion of the TAC has been consistent over time. Pollock roe production remains at a low level but increased over 2022 (Figure 8).
The summer-fall fishing conditions for 2023 were similar to 2022 (Figure 6). The number of hours the fleet required to catch the same tonnage of pollock was also improved relative to 2020. In the B-season catches in the northwestern area increased relative to the previous two years (Figure 9). We updated our work on a measure of fleet dispersion: the relative distance or spread of the fishery in space. Briefly, the calculation computes for a given day, the distance between all trawl tows (within and across boats). These distances are then averaged for year and season. Updated to this year, results indicated that in the A-season dispersion increased slightly but for the B-season in 2023, the fleet appeared to be less dispersed than all the other years and since 2000 (Figure 10).
We continued to investigate the tow specific mean weight of fish. These provide a direct mean somatic mass (pollock body weight) for pollock within a tow. The data arise from the sampled total weight (e.g., of several baskets of pollock) divided by the enumerated number of fish in that sample. Such records exist for each tow. Summing these by extrapolated weight of the pollock catch within that tow, and binning by weight increments (here by 50 gram intervals), allows us to obtain some additional fine-scale information on the size trends in the pollock fishery. The annual patterns of these data suggest that the 2023 A-season size was consistent with the expectation of the 2018 year class predominating the catch (Figure 11). However, the 2023 B-season pattern was smaller than expected. Compiling the data by week we show that the fish size was consistent with the pattern of fish being consistently smaller than expected through the B-season (Figure 12).
The catch of EBS pollock has averaged 1.26 million t in the period since 1979. The lowest catches occurred in 2009 and 2010 when the limits were set to 0.81 million t due to stock declines (). The recent 5-year average (2019-2023) catch has been 1.304 million t. Pollock catches that are retained or discarded (based on NMFS observer estimates) in the Eastern Bering Sea and Aleutian Islands for 1991–2024 are shown in . Since 1991, estimates of discarded pollock have ranged from a high of 9.6% of total pollock catch in 1991 to recent lows of around 0.6% to 1.2%. These low values reflect the implementation of the NMFS’ Improved Retention /Improved Utilization program. Prior to the implementation of the American Fisheries Act (AFA) in 1999, higher discards may have occurred under the “race for fish” and pollock marketable sizes were caught incidentally. Since implementation of the AFA, the vessel operators have more time to pursue optimal sizes of pollock for market since the quota is allocated to vessels (via cooperative arrangements). In addition, several vessels have made gear modifications to avoid retention of smaller pollock. In all cases, the magnitude of discards counts as part of the total catch for management (to ensure the TAC is not exceeded) and within the assessment. Bycatch of other non-target, target, and prohibited species is presented in the section titled Ecosystem Considerations below. In that section it is noted that the bycatch of pollock in other target fisheries is more than double the bycatch of other target species (e.g., Pacific cod) in the pollock fishery.
Management measures
The EBS pollock stock is managed by NMFS regulations that provide limits on seasonal catch. The NMFS observer program data provide near real-time statistics during the season and vessels operate within well-defined limits. In most years, the TACs have been set well below the ABC value and catches have stayed within these constraints ). Allocations of the TAC split first with 10% to western Alaska communities as part of the Community Development Quota (CDQ) program and the remainder between at-sea processors and shore-based sectors. For a characterization of the CDQ program see Haynie (2014). Seung and Ianelli (2016) combined a fish population dynamics model with an economic model to evaluate regional impacts.
Due to concerns that groundfish fisheries may impact the rebuilding of the Steller sea lion population, a number of management measures have been implemented over the years. Some measures were designed to reduce the possibility of competitive interactions between fisheries and Steller sea lions. For the pollock fisheries, seasonal fishery catch and pollock biomass distributions (from surveys) indicated that the apparent disproportionately high seasonal harvest rates within Steller sea lion critical habitat could lead to reduced sea lion prey densities. Consequently, management measures redistributed the fishery both temporally and spatially according to pollock biomass distributions. This was intended to disperse fishing so that localized harvest rates were more consistent with estimated annual exploitation rates. The measures include establishing: 1) pollock fishery exclusion zones around sea lion rookery or haulout sites; 2) phased-in reductions in the seasonal proportions of TAC that can be taken from critical habitat; and 3) additional seasonal TAC releases to disperse the fishery in time.
Prior to adoption of the above management measures, the pollock fishery occurred throughout each of the three major NMFS management regions of the North Pacific Ocean: the Aleutian Islands (1,001,780 km\(^2\) inside the EEZ), the Eastern Bering Sea (968,600 km\(^2\)), and the Gulf of Alaska (1,156,100 km\(^2\)). The marine portion of Steller sea lion critical habitat in Alaska west of 150\(^{\circ}\)W encompasses 386,770 km\(^2\) of ocean surface, or 12% of the fishery management regions.
From 1995–1999 84,100 km\(^2\), or 22% of the Steller sea lion critical habitat was closed to the pollock fishery. Most of this closure consisted of the 10 and 20 nm radius all-trawl fishery exclusion zones around sea lion rookeries (48,920 km\(^2\), or 13% of critical habitat). The remainder was largely management area 518 (35,180 km\(^2\), or 9% of critical habitat) that was closed pursuant to an international agreement to protect spawning stocks of central Bering Sea pollock. In 1999, an additional 83,080 km\(^2\) (21%) of critical habitat in the Aleutian Islands was closed to pollock fishing along with 43,170 km\(^2\) (11%) around sea lion haulouts in the GOA and Eastern Bering Sea. In 1998, over 22,000 t of pollock were caught in the Aleutian Island region, with over 17,000 t taken within critical habitat region. Between 1999 and 2004 a directed fishery for pollock was prohibited in this region. Subsequently, 210,350 km\(^2\) (54%) of critical habitat in the Aleutian Islands was closed to the pollock fishery. In 2000 the remaining phased-in reductions in the proportions of seasonal TAC that could be caught within the BSAI Steller sea lion Conservation Area (SCA) were implemented.
On the EBS shelf, an estimate (based on observer at-sea data) of the proportion of pollock caught in the SCA has averaged about 44% annually. During the A-season, the average is also about 44%. Nonetheless, the proportion of pollock caught within the SCA varies considerably, presumably due to temperature regimes and the relative population age structure. The annual proportion of catch has ranged from an annual low of 11% in 2010 to high of 60% in 1998–the 2019 annual value was 58% and quite high again in the A-season (68%). The higher values in recent years were likely due to good fishing conditions close to the main port. The recent transition from at-sea observer sampling of many catcher vessels to a combination of at-sea electronic monitoring and shore-based observer sampling has resulted in a temporary hiatus in to associate catches with specific areas. Work has progressed to link the position information to offloads so that haul records could be used to evaluate fishing patterns.
The AFA reduced the capacity of the catcher/processor fleet and permitted the formation of cooperatives in each industry sector by the year 2000. Because of some of its provisions, the AFA gave the industry the ability to respond efficiently to changes mandated for sea lion conservation and salmon bycatch measures. Without such a catch-share program, these additional measures would likely have been less effective and less economical (Strong and Criddle (2014)).
An additional strategy to minimize potential adverse effects on sea lion populations is to disperse the fishery throughout more of the pollock range on the Eastern Bering Sea shelf. While the distribution of fishing during the A-season is limited due to ice and weather conditions, there appears to be some dispersion to the northwest area (Figure 5).
The majority (about 56%) of Chinook salmon caught as bycatch in the pollock fishery originate from western Alaskan rivers. This was updated at the June 2022 Council meeting and is activities are monitored and reported closely at the Council (at this website). In summary, additional Chinook salmon bycatch management measures went into effect in 2011 which imposed revised prohibited species catch (PSC) limits. These limits, when reached, close the fishery by sector and season (Amendment 91 to the BSAI Groundfish Fishery Management Plan (FMP) resulting from the NPFMC’s 2009 action). Previously, all measures for salmon bycatch imposed seasonal area closures when PSC levels reached the limit (fishing could continue outside of the closed areas). The current program imposes a dual cap system by fishing sector and season. A goal of this system was to maintain incentives to avoid bycatch at a broad range of relative salmon abundance (and encounter rates). Participants are also required to take part in an incentive program agreement (IPA). These IPAs are approved and reviewed annually by NMFS to ensure individual vessel accountability. The fishery has been operating under rules to implement this program since January 2011.
Further measures to reduce salmon bycatch in the pollock fishery were developed and the Council took action on Amendment 110 to the BSAI Groundfish FMP in April 2015. These additional measures were designed to add protection for Chinook salmon by imposing more restrictive PSC limits in times of low western Alaskan Chinook salmon abundance. This included provisions within the IPAs that reduce fishing in months of higher bycatch encounters and mandate the use of salmon excluders in trawl nets. These provisions were also included to provide more flexible management measures for chum salmon bycatch within the IPAs rather than through regulatory provisions implemented by Amendment 84 to the FMP. The new measure also included additional seasonal flexibility in pollock fishing so that more pollock (proportionally) could be caught during seasons when salmon bycatch rates were low. Specifically, an additional 5% of the pollock can be caught in the A-season (effectively changing the seasonal allocation from 40% to 45% (as noted above in the discussion assosciated with Figure 7). These measures are all part of Amendment 110 and a summary of this and other key management measures is provided in .
There are three time/area closures in regulation to minimize herring PSC impacts: Summer Herring Savings Area 1 an area south of 57\(^\circ\)N latitude and between 162\(^\circ\)W and 164\(^\circ\)W longitude from June 15 through July 1st. Summer Herring Savings Area 2 an area south of 56\(^\circ\) 30’ N latitude and between 164\(^\circ\)W and 167\(^\circ\)W longitude from July 1 through August 15. Winter Herring Savings Area an area between 58\(^\circ\) and 60\(^\circ\)N latitude and between 172\(^\circ\)W and 175\(^\circ\)W longitude from September 1st through March 1st of the next fishing year.
Data
The following lists the data used in this assessment:
Note the 2020 acoustic survey data based on unmanned surface vessel (USV) transects and age-specific proportions were unavailable in this year
Fishery
Catch
Biological sampling by scientifically trained observers form the basis of a major data component of this assessment (as evaluated in Barbeaux et al. 2005). The catch-at-age composition was estimated using the methods described by Kimura (1989) and modified by Dorn (1992). Length-stratified age data are used to construct age-length keys for each stratum and sex. These keys are then applied to randomly sampled catch length frequency data. The stratum-specific age composition estimates are then weighted by the catch biomass within each stratum to arrive at an overall age composition for each year. Data were collected through shore-side sampling and at-sea observers (Barbeaux et al. (2005)). The three strata for the EBS were: i) January–June (all areas, but mainly east of 170\(^\circ\)W); ii) INPFC area 51 (east of 170\(^\circ\)W) from July–December; and iii) INPFC area 52 (west of 170\(^\circ\)W) from July–December. This method was used to derive the age compositions from 1991–2023 (the period for which all the necessary information is readily available). Prior to 1991, we used the same catch-at-age composition estimates as presented in Wespestad et al. (1996).
The catch-at-age estimation method uses a two-stage bootstrap re-sampling of the data. Observed tows were first selected with replacement, followed by re- sampling actual lengths and age specimens given that set of tows. This method allows an objective way to specify the starting values for the input sample size for fitting fishery age composition data within the assessment model. In addition, estimates of stratum-specific fishery mean weights-at-age (and variances) are provided which are useful for evaluating general patterns in growth and growth variability. For example, James. Ianelli, and Williamson (2007) showed that seasonal aspects of pollock condition factor could affect estimates of mean weight-at-age. They showed that within a year, the condition factor for pollock varies by more than 15%, with the heaviest pollock caught late in the year from October-December (although most fishing occurs during other times of the year) and the thinnest fish at length tending to occur in late winter. They also showed that spatial patterns in the fishery affect mean weights, particularly when the fishery is shifted more towards the northwest where pollock tend to be smaller at age. Grüss et al. (2021) showed cold-pool-extent impacts on the spatial map of summer condition and relating environmental conditions to fish condition continues to be an active area of research.
In 2011 the winter fishery catch consisted primarily of age 5 pollock (the 2006 year class) and later in that year age 3 pollock (the 2008 year class) were present. In 2012–2016 the 2008 year class was prominent in the catches with 2015 showing the first signs of the 2012 year-class as three year-olds in the catch (Figure 14; ). However, by 2017 the 2013 year-class began to be also evident and surpassed the 2012 year-class in dominance and persist through to 2021. The unusual pattern of switching adjacent year-classes was examined in 2021 to see if there was a pattern of spatial differences. There was a distinct spatial distribution of the different year-classes. Having adjacent strong year-classes appears to be a new characteristic of the stock. In 2020, an unusual presence of age-2 pollock appeared in the catch, along with some from the 2014 year-class while the 2012 year-class was a smaller part of the catch (Figure 14). By 2021 and 2022, the predominance of 3- and 4-year olds in the catch confirms the abundance year-class from 2018. We note that the center of locations of the 2018 year-class, as plotted based on the locales of samples from that cohort, appears to be more oriented to the south east (by age) when compared to another abundant year-class (the 2008; Figure 15).
The sampling effort for age determinations, weight-length measurements, and length frequencies is shown in , , and . Sampling for pollock lengths and ages by area has been shown to be relatively proportional to catches. The precision of total pollock catch biomass is considered high with estimated CVs to be on the order of 1% (Miller (2005)).
Scientific research catches are reported to fulfill requirements of the Magnuson-Stevens Fisheries Conservation and Management Act. The annual estimated research catches (1963–2022) from NMFS surveys in the Bering Sea and Aleutian Islands Region are given in (). Since these values represent extremely small fractions of the total removals (about 0.02%) they are ignored for assessment purposes.
Surveys
Bottom trawl survey (BTS)
Trawl surveys have been conducted annually by the AFSC to assess the abundance of crab and groundfish in the Eastern Bering Sea since 1979 and since 1982 using standardized gear and methods. For pollock, this survey has been instrumental in providing an abundance index and information on the population age structure. This survey is complemented by the acoustic trawl (AT) surveys that sample mid-water components of the pollock stock. Between 1991 and 2023 the BTS biomass estimates ranged from 2.28 to 8.39 million t () for the design-based estimates). The values used for the assessment (VAST index, see Section 16 for details) are shown in Figure 16. In the mid-1980s and early 1990s several years resulted in above-average biomass estimates. The stock appeared to be at lower levels during 1996–1999 then increased moderately until about 2003 and since then has averaged just over 4 million t (from the standard EBS region using design-based estimators).
These surveys also provide consistent measurements of environmental conditions, such as the sea surface and bottom temperatures. Large-scale zoogeographic shifts in the EBS shelf documented during a warming trend in the early 2000s were attributed to temperature changes (e.g., Mueter and Litzow (2008)). However, after the period of relatively warm conditions ended in 2005, the next eight years were mainly below average, indicating that the zoogeographic responses may be less temperature-dependent than they initially appeared (Kotwicki and Lauth (2013)). Bottom temperatures increased in 2011 to about average from the low value in 2010 but declined again in 2012–2013. In the period 2014–2016, bottom temperatures increased and reached a new high in 2016. In 2018 bottom temperatures were nearly as warm (after 2017 was slightly above average) but was highly unusual due to the complete lack of “cold pool” (i.e., a defined area where water near bottom was less than zero degrees. In 2019, the mean bottom temperature was the warmest during the period the survey has occurred (since 1982; Figure 17). In 2022 and 2023, the bottom temperatures have declined but remain above average.
The AFSC has expanded the area covered by the bottom trawl survey over time. In 1987 the “standard survey area” comprising 6 main strata was increased farther to the northwest and covered in all subsequent years. These two northern strata have varied in estimated pollock abundance. In 2023 about 10% of the pollock biomass was found in these strata compared to a long term average of 5% (). Importantly, this region is contiguous with the Russian border and the NBS region, and treatment of the extent stock shifts between regions continues (e.g., C. A. O’Leary et al. (2021)).
After the increase in 2022, the 2023 survey estimate is similar to the 2021 value and is about 72% of the long term mean. The 2023 pollock density by station appeared to be lower overall with some slight increases on the outer shelf area to the northwest (Figure 18). The VAST model provides density-weighted population shifts in distribution. This can be expressed in north-south and east-west trends over time. A representation of such center of gravity estimates indicate that the stock has moved steadily north since the mid 2000s, but last year shifted south. This year it has shifted back north to some degree (Figure 19). The stock center of gravity also moved east from 2010 to about 2017, then shifted west and seems about at it’s long term mean.
The BTS abundance-at-age estimates show variability in year-class strengths with substantial consistency over time (Figure 20). The abundance of 5-year old pollock (the 2018 year-class) dropped from 2022, but still represents the most abundant year class. The abundance of age-1 pollock in 2023 appears to be about average.
Pollock above 40 cm in length generally appear to be fully selected and in some years, many 1-year olds occur on or near the bottom (with modal lengths around 10–19 cm). Generally speaking, age 2 or 3 pollock (lengths around 20–29 cm and 30–39 cm, respectively) are relatively rare in this survey because they tend to be more pelagic as juveniles. Compared to recent years, the size compositions were consistent with the mid-range categories and consistent with the age data (Figure 21).
Observed fluctuations in survey estimates may be attributed to a variety of sources including unaccounted-for variability in natural mortality, survey catchability, and horizontal migrations and vertical availability (Cole C. Monnahan et al. (2021); Cecilia A. O’Leary et al. (2022)). As an example, some strong year classes appear in the surveys over several ages (e.g., the 1989 year class) while others appear only at older ages (e.g., the 1992 and 2008 year class). Sometimes, initially strong year classes appear to wane in successive assessments (e.g., the 1996 year class estimate (at age 1) dropped from 43 billion fish in 2003 to 32 billion in 2007 (James. Ianelli, and Williamson (2007))). Retrospective analyses (e.g., Parma (1993)) have also highlighted these patterns, as presented in Ianelli et al. (2006, 2011). Kotwicki and Lauth (2013) also found that the catchability of either the BTS or AT survey for pollock is variable in space and time because it depends on environmental variables, and is density-dependent in the case of the BTS survey.
The 2023 survey age compositions were developed from age-structures collected during the survey (June-July) and processed at the AFSC labs within a few weeks after the survey was completed. The level of sampling for lengths and ages in the BTS is shown in . The estimated numbers-at- age from the BTS for strata 1–9 (except for 1982–84 and 1986, when only strata 6 were surveyed) are presented in (based on the method in Kotwicki, Ianelli, and Punt (2014) and then using VAST–see Section 16 for those details). Compared to the previous design-based age composition estimates, those derived from the spatio-temporal model were generally very similar (Figure 22).
In the previous assessments, the BTS mean body mass-at-ages was computed based on the sex-specific mean length-at-age in each year and converted to weight using sex-specific length-weight parameters that were estimated from data prior to 1999. In reconsidering this approach, data on weight-at-age from intervening years have become available and some new methods applied including those corrected by spatio-temporal modeling (Indivero et al. (2023)). This work was adopted in 2022 and values used are shown in . The time series of BTS survey indices is shown in .
The NBS survey area was sampled in 2010, 2017, 2018 (limited to 49 stations), 2019, and 2021-2023. Given that the pollock abundance was quite high in 2017 and 2018, a method for incorporating this information as part of the standard survey was desired. One approach for constructing a full time series that includes the NBS area is to use observed spatial and temporal correlations. We used the vector-autoregressive spatial temporal (VAST) model of Thorson (2019) together with the density-dependent corrected CPUE values from each station (including stations where pollock were absent; ). Please refer to the Section 16 for further details on the implementation. The appendix also includes results that indicate the VAST model diagnostics are reasonable and provide consistent interpretations relative to the observations. Notably, results indicate increased uncertainty in years and areas when stations were missing. As noted in past assessments, application of this index within the stock assessment model required accounting for the time-series covariance estimate.
To date, given other commitments, work on comparing the age-and-growth from NBS samples has stalled. We hope to evaluate these data when they become available in the near future to look at maturity and growth conditions from this region.
Acoustic trawl surveys
Acoustic trawl surveys are typically conducted every other year and are designed to estimate the off- bottom component of the pollock stock (compared to the BTS which are conducted annually and provide an abundance index of the near-bottom pollock). Estimated pollock biomass for the EBS shelf has averaged over 3.2 million t since the time-series was revised to include the water column to 0.5 m (from the historical midwater pollock index to 3 m off bottom) starting in 1994 (). The early 2000s (a relatively ‘warm’ period) were characterized by low pollock recruitment, which was subsequently reflected in lower pollock biomass estimates between 2006 and 2012 (a ‘cold’ period; T. Honkalehto and McCarthy (2015)). In 2014 and 2016 (another ‘warm’ period) with the growth of the strong 2012 year class, AT biomass estimates increased to over 4 million t (). The number of trawl hauls, lengths, and ages sampled from the AT survey are presented in . These surveys have also provided insight on the relative abundance of pollock in areas considered critical to Steller sea lions (the “SCA”; ).
Pollock midwater abundance and distribution were last assessed in 2022. In addition to the traditional (core) survey area, a region north of most transects (the northern extension) was surveyed. Transect spacing, typically 20 nmi, ranged from 40 nmi in the east and middle shelf to 20 nmi in the western shelf due to ship staffing constraints and consequent survey schedule uncertainties.
The 2022 estimated amount of pollock in the core survey area was 9.67 billion fish with a biomass of 3.834 million metric tons (t), just over a 50% increase from the estimate of 5.55 billion fish with a biomass of 2.497 million t in 2018. This was a 6% increase over the 3.617 million t estimated in 2020 by the acoustics-only Saildrone survey. Preliminary population age estimates from 2022 using BTS ages were revised in 2023 using ages from the AT survey. Four-year-olds (2018 year class) dominated the pollock population numbers in the core survey area () comprising 71% of the core area biomass followed by 3-year-olds (2019 year class, 7.6 % of the core area biomass). Slightly more than one-half million t (0.539 million t) of pollock were observed distributed sparsely along the northern extension transects, 12% of the shelf-wide total. Eight year-olds (2014 year class) were the dominant aged pollock in the northern extension (29% by biomass), followed by 7 year-olds (17% of the northern extension biomass) and 4 year-olds (15% of the northern extension biomass).
Relative estimation errors for the total biomass were derived from a one-dimensional (1D) geostatistical method, which accounts for observed spatial structure for sampling along transects (Petitgas (1993), Walline (2007), Williamson and Traynor (1996)). The 2022 relative estimation error for the core survey area was 0.068, slightly higher than the time series mean of 0.043, likely due to increased transect spacing. As in previous assessments, the other sources of error (e.g., target strength, trawl selectivity) were accounted for by inflating the annual error estimates to have an overall average CV of 20% for application within the assessment model. This value was consistent with past model fitting procedures (i.e., the standard deviation of the normalized residuals was very close to 1.0).
Other time series used in the assessment
Japanese fishery CPUE index
An available time series relating the abundance of pollock during the period 1965–1976 was included. This series is based on Japanese fishery catch rates which used the same size class of trawl vessels as presented in Low and Ikeda (1980). In lieu of an objective estimate, we applied a default coefficient of variation of 20% to these data
Biomass index from Acoustic-Vessels-of-Opportunity (AVO)
Acoustic backscatter data (Simrad ES60, 38 kHz) were collected aboard two fishing vessels chartered for the AFSC summer 2024 bottom trawl surveys (F/V Alaska Knight, F/V Northwest Explorer). We processed these Acoustic Vessels of Opportunity (AVO) data each year since 2006 to provide an index of age-1+ midwater pollock abundance. As with last year, we implemented a new subsampling methodology (Levine and De Robertis (2019)) to generate a more spatially extensive AVO index. In developing the new index, we analyzed a 10% systematic subsample of the BTS backscatter data throughout the typical ATS geographic footprint. The new methods were applied to reanalyze years 2009, 2010, 2012, 2014-2019, and 2021-2024 For the remaining 5 years of the time series, the original AVO index (T. Honkalehto et al. (2011), Stienessen et al. (2020)) was rescaled to match the mean of the new AVO time series (Figure 24). For 2024 the AVO data were processed (Lauffenburger et al. (2024)) to provide an index of age-1+ midwater pollock abundance in each year. These pre-publication results are given below (noting that the final results are unlikely to change):
The 2024 AVO index of midwater pollock abundance on the EBS shelf was 2.01 million t, which decreased 19% from 2023 and 31% from 2022. Although not the lowest estimate in the time series, it is the lowest estimate since 2014. This compares with the 2024 AFSC biennial acoustic-trawl survey (ATS) conducted using NOAA Ship Oscar Dyson decreased 25% from 2022.
The correlation between the AVO index and the AT survey biomass remained the same (r^2= 0.895, n= 9 surveys).
The distribution of pollock backscatter east and west of the Pribilof Islands was average since 2009 (33%).
The strongest pollock backscatter during the AVO and AT surveys was measured along the southern portion of the EBS shelf (Figure 4). The center of gravity estimate for the 2024 AVO survey was similar to that of the AVO 2022 and 2023 estimates, whereas the center of gravity estimate for the 2024 AT survey was shifted southeast compared to the AT 2022 estimate, and more similar to the AVO 2022-2024 center of gravity estimates (Figure 5).
Relative to the original index, the correlation between the AVO index and the AT survey biomass was higher (R2 = 0.9, compared to R2 = 0.6 for the same seven ATS-BTS years from the original index). The new and rescaled index trend dropped 15% relative to 2022 but still is 13% above the long-term mean (Figure 24, ; note that the relative error is based on a variance estimation from Petitgas (1993), while the final magnitude of the error term was ascertained based on other model components via an iterative re-weighting process as noted in Ianelli (2023)). The densest spatial distribution of pollock backscatter was predominantly measured along the southern portion of the EBS shelf in the northwest half of the index area (Figure 25). The three grid cells (20 by 20 nautical mile grids used for annual bottom-trawl survey stations) having the strongest pollock backscatter were 2o south of St. Matthew Island close to (58oN, 172oW) and attributed to dense midwater pollock aggregations.
Analytic approach
General model structure
We used a statistical age-structured assessment model conceptually outlined in D. Fournier and Archibald (1982) and extended (e.g., Methot (1990)). This was developed as an appendix to Wespestad et al. (1996) with current specifications presented in the ?@sec-model (J. N. Ianelli and Fournier (1998)). The model was written in ADMB—a library for non-linear estimation and statistical applications (David a. Fournier et al. (2012)). The data updated from last year’s analyses include:
The 2023 fishery age composition data
The catch biomass estimates through the current year
The 2024 bottom-trawl survey index, weight, and age composition data
The 2023 acoustic-trawl age composition data were revised using only samples collected from that survey (previously the age compositions were estimated using the bottom-trawl survey age-length keys)
A completely revised time series of AVO backscatter data collected opportunistically from the bottom trawl survey.
A simplified version of the assessment (with mainly the same data and likelihood-fitting method) is included as a supplemental multi-species assessment model. As presented since 2016, it allows for trophic interactions among key prey and predator species and for pollock, and it can be used to evaluate age and time-varying natural mortality estimates in addition to alternative catch scenarios and management targets (see this volume: EBS multi-species model).
Description of alternative models
In the 2019 assessment, the spatio-temporal model fit to BTS CPUE data including stations from the NBS was expanded using the VAST methods detailed in Thorson (2018). This data treatment was included as a model alternative and adopted for ABC/OFL specifications by the SSC in 2020 along with other modifications including a spatio-temporal treatment of the age composition data. This year, we examined additional model and data modifications as presented in J. Ianelli (2023).
By the SSC’s numbering scheme, last year’s model was designated Model 20.0, which here we contrast with a revised model 23.0. As usual, we also provide an incrememntal evaluation of the influence of new data introduced in 2023 (here as applied to Model 23.0).
As noted in J. Ianelli (2023), we continue to provide some facility to test different stock assessment software (as noted in Li et al. (2021)).
Input sample size
Sample sizes for age-composition data were re-evaluated in J. Ianelli (2023) and found to be consistent with the relative variability allowed for selectivities and with the observation errors specified for the indices. Principally, this work resulted in tuning the recent era (1991-present year) to an average sample sizes of 350 for the fishery and then using estimated values for the period 1978-1990 and as earlier (). As rationalized in earlier assessments, we found that assuming average values of 100 and 50 for the BTS and ATS data, respectively resulted in consistent model fits and were (relatively) appropriate given the sampling levels among these surveys. The inter-annual variability reflects the variability in the number of hauls sampled for ages in the ATS data. For the BTS data we adopted the results presesnted in Hulson et al. (2023). We re-evaluated tuning following Francis (2011) (equation TA1.8).
Recent work has shown ways to improve estimation schemes that deal with the interaction between flexibility in fishery selectivity and statistical properties of composition data sample size. Specifically, the Dirichlet-multinomial using either Laplace approximation (Thorson, Hicks, and Methot (2015)) or adnuts (C. C. Monnahan and Kristensen (2018)) should be implemented (e.g., as shown by Xu, Thorson, and Methot (2020)). Progress in 2023 has lagged on this, but with the advent of some alternative three dimensional mixed-effects approaches to weight-at-age and selectivity (Cheng et al. (2023)), development of more elaborate approaches are being pursued.
Parameters estimated outside of the assessment model
Natural mortality and maturity at age
The baseline model specification has been to use constant natural mortality rates at age (M=0.9, 0.45, and 0.3 for ages 1, 2, and 3+ respectively (Wespestad and Terry (1984)). When predation was explicitly considered estimates tended to be higher and more variable (Holsman et al. this volume; Holsman and Aydin (2015); Livingston and Methot (1998); Hollowed, Ianelli, and Livingston (2000)). Clark et al. (1999) found that specifying a conservative (lower) natural mortality rate may be advisable when natural mortality rates are uncertain. More recent studies confirm this (e.g., Johnson et al. (2014)).
In the supplemental multi-species assessment model alternative values of age and time-varying natural mortality are presented. As in past years the estimates indicate higher values than used here. In the 2018 assessment we evaluated natural mortality, and it was noted that the survey age compositions favored lower values of M while the fishery age composition favored higher values. This is consistent with the patterns seen in the BTS survey data as they show increased abundances of “fully selected” cohorts. Hence, given the model specification (asymptotic selectivity for the BTS age composition data), lower natural mortality rates would be consistent with those data. Given these trade-offs, structural model assumptions were held to be the same as previous years for consistency (i.e., the mortality schedule presented below).
Maturity-at-age values used for the EBS pollock assessment were originally based on Smith (1981) and were later reevaluated via histological methods (e.g., J. Stahl (2004); J. P. Stahl and Kruse (2008), J. N. Ianelli (2005)). These studies found year-class effects and some inter-annual variability but general consistency with the original schedule of proportion mature at age.
With respect to assumptons about natural mortality, we evaluated applying results from an adjacent stock (Ianelli James N. and McKelvey (2022)) in the 2022 assessment. We found the results were consistent with past assumptions therefore again applyed the following age-specific values for M (Smith (1981)) and maturity-at-age:
Length and weight-at-age
Age determination methods have been validated for pollock (Kimura et al. (1992), Kimura et al. (2006), and Kastelle and Kimura (2006)). EBS pollock size-at-age show important differences in growth with differences by area, year, and year class. Pollock in the northwest area are typically smaller at age than pollock in the southeast area. The differences in average weight-at-age are taken into account by stratifying estimates of catch-at-age by year, area, season, and weighting estimates proportional to catch.
The assessment model for EBS pollock accounts for numbers of individuals in the population. As noted above, management recommendations are based on allowable catch levels expressed as tons of fish. While estimates of pollock catch-at-age are based on large data sets, the data are only available up until the most recent completed calendar year of fishing (e.g., 2022 for this year). Consequently, estimates of weight-at-age in the current year are required to map total catch biomass (typically equal to the quota) to numbers of fish caught (in the current year). Therefore, if there are errors (or poorly accounted uncertainty) in the current and future mean weight-at-age, this can translate directly into errors between the expected fishing mortality and what mortality occurs. For example, if the mean weight-at-age is biased high, then an ABC (and OFL) value will result in greater numbers of fish being caught (and fishing mortality being higher due to more fish fitting within the ABC).
As in previous assessments, we explored patterns in size-at-age and fish condition. Using the NMFS fishery observer data on weight given length we:
extracted all data where non-zero measurements of pollock length and weight were available between the lengths of 35 and 60 cm for the EBS region
computed the mean value of body mass (weight) for each cm length bin over all areas and time
divided each weight measurement by that mean cm-specific value (the “standardization” step)
plotted these standardized values by different areas, years, months etc. to evaluate condition differences (pooling over ages is effective as there were no size-specific biases apparent)
In the first instance, the overarching seasonal pattern in body mass relative to the mean shows that as the winter progresses prior to peak spawning, pollock are generally skinnier than average whereas in July, the median is about average (Figure 26). As the summer/fall progresses, fish were at their heaviest given length (Figure 26). This is also apparent when the data are aggregated by A- and B-seasons (and by east and west of 170\(^\circ\)W; referred to as SE and NW respectively) when plotted over time (Figure 27, where stratum 1 = A season, stratum 2 = B season SE, and stratum 3 = B season NW). Combining across seasons, the fishery data shows that recent years were below average weight given length (Figure 28 ; note that the anomalies are based on the period 1991-2023).
Examining the weight-at-age, there are also patterns of variability that vary due to environmental conditions in addition to spatial and temporal patterns of the fishery. Based on the bootstrap distributions and large sample sizes, the within-year sampling variability for pollock is small. However, the between-year variability in mean weights-at-age is relatively high (). The coefficients of variation between years are on the order of 6% to 9% (for the ages that are targeted) whereas the sampling variability is generally around 1% or 2%. The approach to account for the identified mean weight-at-age having clear year and cohort effects was continued (e.g., Figure 29). Details were provided in appendix 1A of Ianelli et al. (2016). The results from this method showed the relative variability between years and cohorts and provide estimates for 2024–2026 (). How these fishery weights-at-age estimates can be supplemented using survey weights-at-age is further illustrated in Figure 30.
In the 2020 and 2021 fishery, the average weight-at-age for ages 6-8 (the 2012-2014 year classes) was below the time series average. These cohorts have fluctuated around their means in recent years (Figure 29). To examine this more closely, we split the bootstrap results into area-season strata and were able to get an overall picture of the pattern by strata (Figure 31 and Figure 32). This showed that the mean weight-at-age is higher in the the B-season in the area east of 170\(^\circ\)W compared to the A-season and B-season in the area west of 170\(^\circ\)W.
Parameters estimated within the assessment model
For the selected model, 1366 parameters were estimated conditioned on data and model assumptions. Initial age composition, subsequent recruitment, and stock- recruitment parameters account for 80 parameters. This includes vectors describing the initial age composition (and deviation from the equilibrium expectation) in the first year (as ages 2–15 in 1964) and the recruitment mean and deviations (at age 1) from 1964–2024 and projected recruitment variability (using the variance of past recruitments) for five years (2025–2030). The two- parameter stock-recruitment curve (see ?@sec-model) is included in addition to a term that allows the average recruitment before 1964 (that comprises the initial age composition in that year) to have a mean value different from subsequent years. Note that the stock-recruit relationship is fit only to stock and recruitment estimates from 1979 year-class through to the 2022 year-class.
Fishing mortality is parameterized to be semi-separable with year and age (selectivity) components. The age component is allowed to vary over time; changes are allowed in each year. The mean value of the age component is constrained to equal one and the last 5 age groups (ages 11–15) are specified to be equal. This latter specification feature is intended to reduce the number of parameters while acknowledging that pollock in this age-range are likely to exhibit similar life-history characteristics (i.e., unlikely to change their relative availability to the fishery with age). The annual components of fishing mortality result in 61 parameters and the age-time selectivity schedule forms a 10x61 matrix of 610 parameters bringing the total fishing mortality parameters to 671. The rationale for including time- varying selectivity has recently been supported as a means to improve retrospective patterns (Szuwalski et al. 2017) and as best practice (Martell and Stewart (2013)).
For surveys and indices, the treatment of the catchability coefficient, and interactions with age-specific selectivity require consideration. For the BTS index, selectivity-at-age is estimated with a logistic curve in which year specific deviations in the parameters is allowed. Such time-varying survey selectivity is estimated to account for changes in the availability of pollock to the survey gear and is constrained by pre-specified variance terms. Presently, these variance terms have been set based on balancing input data-based variances and are somewhat subjective. For the AT survey, which originally began in 1979 (the current series including data down to 0.5 m from bottom begins in 1994), optional parameters to allow for age and time-varying patterns exist but for this assessment and other recent assessments, ATS selectivity is constant over time. Overall, four catchability coefficients were estimated: one each for the early fishery catch-per-unit effort (CPUE) data (from Low and Ikeda, 1980), the VAST combined bottom trawl survey index, the AT survey data, and the AVO data. An uninformative prior distribution is used for all of the indices. The selectivity parameters for the 2 main indices (BTS and ATS) total 336 (the CPUE and AVO data mirror the fishery and AT survey selectivities, respectively).
Additional fishing mortality rates used for recommending harvest levels are estimated conditionally on other outputs from the model. For example, the values corresponding to the \(F_{40\%}\) \(F_{35\%}\) and \(F_{MSY}\) harvest rates are found by satisfying the constraint that, given age-specific population parameters (e.g., selectivity, maturity, mortality, weight-at-age), unique values exist that correspond to these fishing mortality rates. The likelihood components that are used to fit the model can be categorized as:
- Total catch biomass (log-normal, \(\sigma=0.05\))
- Log-normal indices of pollock biomass; bottom trawl surveys assume annual estimates of sampling error, as represented in Figure 16 along with the covariance matrices (for the density-dependent and VAST index series); for the AT index the annual errors were specified to have a mean CV of 0.20; while for the AVO data, a value a mean CV was tuned for consistency with other data and resulted in a value of 23%).
- Fishery and survey proportions-at-age estimates (multinomial with effective sample sizes presented ).
- Age 1 index from the AT survey (CV set equal to 30% as in prior assessments).
- Selectivity constraints: penalties/priors on age-age variability, time changes, and decreasing (with age) patterns.
- Stock-recruitment: penalties/priors involved with fitting a stochastic stock-recruitment relationship within the integrated model.
- “Fixed effects” terms accounting for cohort and year sources of variability in fishery mean weights-at-age estimated based on available data from 1991-2023 from the fishery (and 1982-2024 for the bottom-trawl survey data) and externally estimated variance terms as described in Appendix 1A of Ianelli et al. (2016; see Figure 30).
Work evaluating temperature and predation-dependent effects on the stock- recruitment estimates continues (Spencer et al. (2016)). This approach modified the estimation of the stock-recruitment relationship by including the effect of temperature and predation mortality. A relationship between recruitment residuals and temperature was noted (similar to that found in Mueter et al. (2011) and subsequently noted in Thorson, Cheng, et al. (2020)) and lower pollock recruitment during warmer conditions might be expected. Similar results relating summer temperature conditions to subsequent pollock recruitment for recent years were also found by Yasumiishi et al. (2015) where research suggests that summer warmth is associated with earlier diapause of copepods (Thorson, Adams, et al. (2020)), such that a fall (but not spring) survey of copepod densities is also associated with cold conditions and elevated recruitment (Eisner et al. (2020)).
Results
Model evaluation
A sequential sensitivity of available new data showed that adding the new data from 2023 had very minor changes and impact on the spawning biomass estimates (Figure 33; top panel). The largest effect on all the changes arose from the revision to the mean body weight-at-age used for the spawning biomass calculations (Figure 33; bottom panel). This was shown in J. Ianelli (2023). Nonetheless, diagnostics of all the changes relative to model fits are given in and a comparison of management quantities for the final base model is given in ).
In the 2020 assessment, SRR evaluations related to Tier 1 classification showed that dropping the influence of the 1978 year-class in the estimation lowered the steepness of the curve and that when the influence of the prior distribution was removed the residual pattern for estimates near the origin was particularly bad (all below the curve). From those results we conclude that the prior specification was appropriate because we place priority on fitting estimated recruits near the slope at the origin better. In the 2021 assessment we showed that conditioning the SRR to fit the condition of having the “actual” \(F_{MSY}\) equal some \(F_{MSY}\) proxies (e.g., equal \(F_{35\%}\)) resulted in more conservative ABCs due to shallower initial slopes. A conclusion from these exercises was that the SPR proxy for \(F_{MSY}\) implies a reasonable “shape” to the SRR.
The fit to the early Japanese fishery CPUE data (Low and Ikeda 1980) was consistent with the estimated population trends for this period (Figure 34). The model fits the fishery-independent index from the 2006–2023 AVO data well through most of the period but the model predicts lower biomass than the index data indicate in 2023 (Figure 35). The model fits to the bottom-trawl survey biomass (the density-dependent corrected series) were reasonable and within the observation error bounds (Figure 36). The model fit to the BTS biomass index predicts fewer pollock than observed in the 2014 and 2015 survey but then varied in subsequent years (Figure 36). The fit to the acoustic-trawl survey biomass series (including the USV data from 2020) was consistent with the specified observation uncertainty (Figure 37).
The estimated parameters and standard errors are provided online. The code for the model (with dimensions and links to parameter names) and input files are available here.
The input sample size (as tuned in 2016 using “Francis Weights”) can be evaluated visually for consistency with expectations of mean annual age for the different gear types (Figure 38; Francis 2011). The estimated selectivity pattern changes over time and reflects to some degree the extent to which the fishery is focused on particularly prominent year-classes (Figure 39). The model fits the fishery age-composition data quite well under this form of selectivity (Figure 40).
Bottom-trawl survey selectivity estimates are shown in Figure 41. The pattern of bottom trawl survey age composition data in recent years shows a decline in the abundance of age 10+ pollock since 2011 (Figure 42). Through the time series of the available data, the model predicted proportions of the 2012 and 2013 year classes varied in terms of under- and over- estimates as the 2013 year-class became more common in the data (Figure 42). The ATS selectivity varies slightly among ages and years (Figure 43). This enhances the fit to the age composition data while still tracking the large year classes through the population (Figure 44).
As in past assessments, we evaluated the multivariate posterior distribution using Monte-Carlo Markov chain (MCMC) simulation methods. This year we adopted the no-uturn sampling approach from ADMB but upgraded and packaged within R (adnuts, C. C. Monnahan and Kristensen (2018)). This allowed thorough sampling diagnostics and was able to sample the posterior efficiently within a few hours (or less). This new package also demonstrated that the asymptotic parameter standard deviations were reasonable approximations of the marginal densities from the integrated posterior distribution (Figure 45). As before, we evaluated how selected parameters relate by doing a pairwise (along with their marginal distributions; Figure 46). This illustrates how key parameters relate to management parameters of interest. For example, the stock recruitment steepness is negatively correlated to the resulting \(B_{MSY}\) estimate. We also compare the point estimates (highest posterior density) with the mean of the posterior marginal distribution of the 2024 spawning biomass. This showed that the point estimate was similar to the mean of the marginal posterior distribution (Figure 47). As an additional part of the Tier 1 consideration, we evaluated the posterior density of \(F_{MSY}\) and is provided in Figure 48 for reference.
We added code for producing posterior predictive distributions (e.g., for the two acoustic indices in Figure 49. Additionally, we developed some preliminary diagnostics to evaluate how the model’s posterior components affect key parameters of interest. For example, it is useful to know the relative impact of the 2018 year-class on the next year’s spawning biomass (Figure 50). Additionally, what different components (in negative log-likelihood terms) conflict or interact with such a critical parameter (Figure 51)
Retrospective analysis
Running the assessment model over a grid with progressively fewer years included (going back to 10 years, i.e., assuming the data extent ended in 2014) results in a fair amount of variability in spawning biomass (Figure 52). Last year with the lower than expected survey biomass estimate followed by an increase this year, the retrospective pattern degraded with an average bias (Mohns \(\rho\) equal to 1300 for the 10 year retrospective).
For the recruitment side, the retrospective pattern shows two key results. First, the 2018 year-class (age 1 recruits in 2019) shows up as a big estimate just this year (Figure 53). Second, the retrospective pattern shows how an equally abundant year-class occurred from the 2012 year-class for three years (with data terminating in 2016, 2017, and 2018). Then, in 2019 and in subsequent years that estimate dropped by over 10% and became the 2012 and the 2013 year-class. In the 2022 assessment we adjusted the value downwards to be equal to the mean of some earlier year classes. This year, we simply accepted the estimate for projections given a better confirmation on the magnitude of the 2018 year class.
Related to this issue of consistency in year-class estimation, and in response to an SSC request, we evaluated how the influence of additional years of data affected year-class estimates. Figure 54 and Figure 55 illustrate how year-class estimates can vary for retrospective analyses. These figures show some of the change in relative abundance between the 2012 and 2013 year-classes and how the 2008 year estimate dissipated some as more data became available.
In response to previous SSC requests to evaluate how selectivity is used for ABC and catch advice, we used the retrospective runs to show how the “projected” selectivity compared with subsequent estimates which had the benefit of more data (?@fig-retro_sel). To explain this figure, and taking the 2023 panel as an example, the blue line in that panel represents the projected estimate from the 2022 “peel” (the current model projecting to 2023 using only data up until 2022). The dots represent estimates from each “peel” and the dots in the 2022 panel are based on this year’s estimated selectivity. In general, the projected selectivity conformed reasonably well with subsequent estimates. To further summarize these results, we also computed a summary statistic as the mean age of selection (independent of any age-specific stock size):
\(\bar a =\frac{\sum{S_a a}}{\sum{S_a}}\)
where \(S_a\) is the selectivity at age (ages 1 to 11). This statistic showed that recently the projection was biased towards younger pollock but earlier on, the bias was toward older fish (Figure 56).
Since selectivity varies over time, and the fact that fishing mortality rates for management advice depend on the assumed future selectivity, we evaluate the pattern of \(F_{MSY}\) rates given different selectivity assumptions (i.e., Figure 39). In the 2020 and 2021 assessment, because of the indications of small pollock being unusually present in the fishery, we chose a selectivity pattern from history that reflected tendency towards younger fish (specifically, that from 2005). Using the statistic on mean selected age, we found that the corresponding \(F_{MSY}\) showed a correlation (Figure 57). This figure reveals how shifts in the relative age of fish selected impact \(F_{MSY}\) estimates.
Time series results
The time series of begin-year biomass estimates (ages 3 and older) suggests that the abundance of Eastern Bering Sea pollock remained at a high level from 1982–88, with estimates ranging from 8 to 12 million t (). Historically, biomass levels increased from 1979 to the mid-1980s due to the strong 1978 and relatively strong 1982 and 1984 year classes recruiting to the fishable population. The stock is characterized by peaks in the mid-1980s, the mid-1990s and again appears to be increasing to a peak of more than 12 million t in 2016 following the low in 2008 of 4.35 million t. The estimate for 2024 is trending downward and at 9.09 million t with 2025 estimated at 8.28 million t.
The level of fishing relative to biomass estimates shows that the spawning exploitation rate (SER, defined as the percent removal of egg production in each spawning year) has been mostly below 20% since 1980 (Figure 58). During 2006 and 2007 the rate averaged more than 20% and the average fishing mortality increased during the period of stock decline. The estimate for 2009 through 2018 was below 20% due to the reductions in TACs relative to the maximum permissible ABC values and increases in the spawning biomass. The fishing mortality has fluctuated since 2010-2015 but, unlike last year’s upward trend, the improved spawning biomass condition has held this rate tending toward lower levels. Age specific fishing mortality rates reflect these patterns and show some increases in the oldest ages from 2011–2013 but relatively stable (Figure 59). The estimates of age 3+ pollock biomass showed a large drop last year compared to several of the earlier years but this has reversed in the current assessment (Figure 60, ).
Estimated numbers-at-age are presented in () and estimated catch-at-age values are presented in (). Estimated summary biomass (age 3+), female spawning biomass, and age-1 recruitment are given in ().
To evaluate past management and assessment performance it can be useful to examine estimated fishing mortality relative to reference values. For EBS pollock, we computed the reference fishing mortality from Tier 1 (unadjusted) and recalculated the historical values for \(F_{MSY}\) (since selectivity has changed over time). Since 1977 the current estimates of fishing mortality suggest that during the early period, harvest rates were above \(F_{MSY}\) until about 1980. Since that time, the levels of fishing mortality have averaged about 35% of the \(F_{MSY}\) level (Figure 61). Projections of spawning stock biomass given the 2025 estimate of fishing mortality rate given catches equal to the 2024 values shows a decline through 2021 and then an increase after; albeit with considerable uncertainty due to uncertainty in recruitment (Figure 62).
Recruitment
Model estimates indicate that the 2008, 2012, 2013, and now the 2018 year classes are above average (Figure 63). The 2018 year class is nearly 4 times bigger than average with a CV of about 16%. The stock-recruitment curve as fit within the integrated model shows the variability of the estimated curve (Figure 64). Note that the 2021 and 2022 year classes (as age 1 recruits in 2022 and 2023) were excluded from the stock-recruitment curve estimation as per convention and guidance from NPFMC. Separate from fitting the stock-recruit relationship within the model, examining the estimated recruits-per-spawning biomass shows variability over time but seems to lack trend and also is consistent with the Ricker stock- recruit relationship used within the model (Figure 65).
Environmental factors affecting recruitment are considered important and contribute to the variability. Previous studies linked strong Bering Sea pollock recruitment to years with warm sea temperatures and northward transport of pollock eggs and larvae (Wespestad et al. 2000; Mueter et al. 2006). As part of the Bering-Aleutian Salmon International Survey (BASIS) project research has also been directed toward the relative density and quality (in terms of condition for survival) of young-of-year pollock. For example, Moss et al. (2009) found age-0 pollock were very abundant and widely distributed to the north and east on the Bering Sea shelf during 2004 and 2005 (warm sea temperature; high water column stratification) indicating high northern transport of pollock eggs and larvae during those years. Mueter et al. (2011) found that warmer conditions tended to result in lower pollock recruitment in the EBS. This is consistent with the hypothesis that when sea temperatures on the eastern Bering Sea shelf are warm and the water column is highly stratified during summer, age-0 pollock appear to allocate more energy to growth than to lipid storage (presumably due to a higher metabolic rate), leading to low energy density prior to winter. This then may result in increased over-winter mortality (Swartzman et al. (2005), Winter, Swartzman, and Ciannelli (2005)). J. N. Ianelli et al. (2011)) evaluated the consequences of current harvest policies in the face of warmer conditions with the link to potentially lower pollock recruitment and noted that the current management system is likely to face higher chances of ABCs below the historical average catches. Also, as part of the evaluation of stationarity given periods of “regimes”, we revisited estimated mean recruitment during different periods previously identified as being unique (Figure 66). This shows that given the revised estimate of the 2018 year class, the impact of the recent warm conditions suggest that the recent period (2000-present) is similar to the mean since 1977.
Harvest recommendations
Status summary
The estimate of \(B_{MSY}\) is 2,352 kt (with a CV of 33%) which is less than the projected 2025 spawning biomass of 2,900 kt; (). For 2025, the estimates put the stock in Tier 1a. The corresponding maximum permissible ABC would thus be 2,868,000 t with a fishable biomass estimated at around 6,984 kt (). For the current year spawning biomass this corresponds to 138% of the \(B_{MSY}\) level. A diagnostic (see ?@sec-model) on the impact of fishing shows that the 2024 spawning stock size is about 58% of the predicted value had no fishing occurred since 1978 ().
The probability that the current stock size is below 20% of \(B_{0}\) (a level important for additional management measures related to Steller sea lion recovery) is <0.1% for 2025 and 2026.
In response to a past request from the SSC, we continued to include results from projections based on Tier 2. We report the “standard” Tier 2 ABC calculation using the point estimate (the mean of the posterior distribution) of \(F_{MSY}\). Therefore, for 2025 the Tier 2a ABC would be 2,628,000 t. Since we have estimates of the harmonic mean (from Tier 1 calculations) an alternative Tier 2 estimate using that in place of the arithmetic mean \(F_{MSY}\) results in an ABC of 2,259,000 t.
In summary, the criterion for Tier 1 depends on a reliable estimate of \(F_{MSY}\) and the uncertainty (the PDF). Tier 2 also requires a reliable estimate of \(F_{MSY}\) (without the PDF requirement). Given the seemingly reasonable posterior marginal density for \(F_{MSY}\), it seems if Tier 1 criterion is unmet, then so would the requirement for Tier 2. Adopting Tier 3, while in principle may result in more conservative catch advice, uses less information available about the stock productivity and requires adopting more assumptions (i.e., that \(F_{35\%}\) is a reasonable proxy for \(F_{MSY}\)). As noted below in the section on risk evaluations, there are reasons for increased concerns. However, these seem to be unrelated to overall stock productivity as relates to the SRR and estimates of \(F_{MSY}\). Consequently, our overall analysis continues to support the SSC’s classification of this stock to be within Tier 1.
Amendment 56 Reference Points
Amendment 56 to the BSAI Groundfish Fishery Management Plan (FMP) defines overfishing level (OFL), the fishing mortality rate used to set OFL (FOFL), the maximum permissible ABC, and the fishing mortality rate used to set the maximum permissible ABC. The fishing mortality rate used to set ABC (\(F_{ABC}\)) may be less than this maximum permissible level, but not greater. Estimates of reference points related to maximum sustainable yield (MSY) are currently available. However, we present both reference points for pollock in the BSAI to retain the option for consideration of either Tier 1, 2, or Tier 3 values from the harvest control rules provided in Amendment 56. These Tiers require reference point estimates for biomass level determinations. Consistent with other groundfish stocks, the following values are based on recruitment estimates from post-1976 spawning events (recognizing the the 1978 year class is excluded from the MSY calculations but included in the SPR calculations):
Specification of OFL and Maximum Permissible ABC
Under Amendment 56 of the BSAI Groundfish FMP, the SSC qualified this stock as satisfying the Tier 1 conditions. As such, the harmonic mean value of \(F_{MSY}\) —here computed as an exploitation rate—is applied to the fishable biomass for computing ABC levels. For details on the risk-averse properties of this approach see Thompson (1996). For a future year, the fishable biomass is defined as the sum over ages of predicted begin-year numbers multiplied by age specific fishery selectivity and estimated mean body mass-at-age. The uncertainty in the average weights-at-age projected for the fishery and “future selectivity” has been demonstrated to affect the buffer between ABC and OFL (computed as 1-ABC/OFL) for Tier 1 maximum permissible ABC (Ianelli et al. 2015). The uncertainty in future mean weights-at-age had a relatively large impact as did the selectivity estimation (see the section above on retrospective behavior and Figure 57).
Since the 2025 female spawning biomass is estimated to be above the \(B_{MSY}\) level (2,352 kt) and above the \(B_{40\%}\) value (2,328 kt) in 2025 and if the 2024 catch is as specified above, then the OFL and maximum permissible ABC values by the different Tier categorizations would be:
Note that the values presented for 2024 assumed a catch of 1,300,000 t in 2023.
Standard Harvest Scenarios and Projection Methodology
A standard set of projections is required for each stock managed under Tiers 1, 2, or 3 of Amendment 56 to the FMP. This set of projections encompasses seven harvest scenarios designed to satisfy the requirements of Amendment 56, the National Environmental Policy Act, and the Magnuson-Stevens Fishery Conservation and Management Act (MSFCMA). While EBS pollock is generally considered to fall within Tier 1, the standard projection model requires knowledge of future uncertainty in \(F_{MSY}\). Since this would require a number of additional assumptions that presume future knowledge about stock-recruit uncertainty, the projections in this subsection are based on Tier 3.
For each scenario, the projections begin with the vector of 2024 numbers at age estimated in the assessment. This vector is then projected forward to the beginning of 2025 using the schedules of natural mortality and selectivity described in the assessment and the best available estimate of total (year-end) catch assumed for 2024. In each subsequent year, the fishing mortality rate is prescribed on the basis of the spawning biomass in that year and the respective harvest scenario. Annual recruits are simulated from an inverse Gaussian distribution whose parameters consist of maximum likelihood estimates determined from the estimated age-1 recruits. Spawning biomass is computed in each year based on the time of peak spawning and the maturity and weight schedules described in the assessment. Total catch is assumed to equal the catch associated with the respective harvest scenario in all years. This projection scheme is run 1,000 times to obtain distributions of possible future stock sizes and catches under alternative fishing mortality rate scenarios.
Five of the seven standard scenarios support the alternative harvest strategies analyzed in the Alaska Groundfish Harvest Specifications Final Environmental Impact Statement. These five scenarios, which are designed to provide a range of harvest alternatives that are likely to bracket the final TAC for 2025, are as follows (“\(maxFABC\)” refers to the maximum permissible value of FABC under Amendment 56):
The latter two scenarios are needed to satisfy the MSFCMA’s requirement to determine whether a stock is currently in an overfished condition or is approaching an overfished condition (for Tier 3 stocks, the MSY level is defined as \(B_{35\%}\)).
Projections and status determination
For the purposes of these projections, we present results based on selecting the \(F_{40\%}\) harvest rate as the \(F_{ABC}\) value and use \(F_{35\%}\) as a proxy for \(F_{MSY}\). Scenarios 1 through 7 were projected 14 years from 2024 ( for Model 23.0–including the 1978 year-class as is convention for Tier 3 estimates). Under catches set to Tier 3 ABC estimates, the expected spawning biomass is well above \(B_{35\%}\) and is expected to be drop below \(B_{40\%}\) by 2026 (given mean recruitment; Figure 67 and assuming catches >2 million t in 2025).
Any stock that is below its minimum stock size threshold (MSST) is defined to be overfished. Any stock that is expected to fall below its MSST in the next two years is defined to be approaching an overfished condition. Harvest scenarios 6 and 7 are used in these determinations as follows:
Is the stock overfished? This depends on the stock’s estimated spawning biomass in 2024:
If spawning biomass for 2024 is estimated to be below 1/2 \(B_{35\%}\) the stock is below its MSST.
If spawning biomass for 2024 is estimated to be above \(B_{35\%}\), the stock is above its MSST.
If spawning biomass for 2024 is estimated to be above 1/2 \(B_{35\%}\) but below \(B_{35\%}\), the stock’s status relative to MSST is determined by referring to harvest scenario 6 ( through ). If the mean spawning biomass for 2034 is below \(B_{35\%}\), the stock is below its MSST. Otherwise, the stock is above its MSST.
Is the stock approaching an overfished condition? This is determined by referring to harvest Scenario 7:
If the mean spawning biomass for 2024 is below 1/2 \(B_{35\%}\), the stock is approaching an overfished condition.
If the mean spawning biomass for 2024 is above \(B_{35\%}\), the stock is not approaching an overfished condition.
If the mean spawning biomass for 2026 is above 1/2 \(B_{35\%}\) but below \(B_{35\%}\), the determination depends on the mean spawning biomass for 2036. If the mean spawning biomass for 2036 is below \(B_{35\%}\), the stock is approaching an overfished condition. Otherwise, the stock is not approaching an overfished condition.
For scenarios 6 and 7, we conclude that pollock is above MSST for the year 2024, and it is expected to be above the “overfished condition” based on Scenario 7 (the mean spawning biomass in 2024 is between the 1/2 \(B_{35\%}\) and \(B_{35\%}\) estimate but by 2036 the stock is above \(B_{35\%}\); (). Based on this, the EBS pollock stock is being fished below the overfishing level and is not approaching an overfished condition.
To fulfill reporting requirements for NOAA’s Species Information System, we computed the average fishing mortality rate corresponding to the specified OFL for the last complete year (2023). This hypothetical 2023 \(F_{OFL}\) from this year’s model was estimated to be 0.262 for EBS pollock (assuming this year’s estimated 2022 selectivity and weight-at-age).
ABC Recommendation
ABC levels are affected by estimates of \(F_{MSY}\) which depend principally on the estimated stock-recruitment steepness parameter, demographic schedules such as selectivity-at-age, maturity, and growth. The current stock size (both spawning and fishable) is estimated to be above average levels and projections indicate the potential for further declines. Updated data and analysis result in an estimate of 2024 spawning biomass (3,260 kt) which is about 138% of \(B_{MSY}\) (2,352 kt). This follows a short period of decline from 2017-2020 followed by a previously unexpected increase due to revised estimates of the 2018 year class. Treating all new data the same way as in the past, this estimate suggests that it would be the biggest year-class on record (79,200 age 1 numbers), but with considerable uncertainty.
Given the same estimated aggregate fishing effort as in 2022 and given the estimated stock trend, the constant-F scenario would yield about 1.3 million t. To obtain 2023 catches on the order of 1.45 million t, given the base model estimates, would require about 18% more effort than what was estimated from 2022.
Should the ABC be reduced below the maximum permissible ABC?
The SSC in its September 2018 minutes recommended that assessment authors and Plan Teams use the risk table below when determining whether to recommend an ABC lower than the maximum permissible. The details of the risk table are provided below. Given the concerns listed there, we recommend reducing the ABC to the value provided under Tier 3 projections.
The table is applied by evaluating the severity of four types of considerations that could be used to support a scientific recommendation to reduce the ABC from the maximum permissible. Examples of the types of concerns that might be relevant include the following (as identified by the work-group):
Assessment considerations
- Data-inputs: biased ages, skipped surveys, lack of fishery-independent trend data
- Model fits: poor fits to fits to fishery or survey data, inability to simultaneously fit multiple data inputs.
- Model performance: poor model convergence, multiple minima in the likelihood surface, parameters hitting bounds.
- Estimation uncertainty: poorly-estimated but influential year classes.
- Retrospective bias in biomass estimates.
- Data-inputs: biased ages, skipped surveys, lack of fishery-independent trend data
Population dynamics considerations—decreasing biomass trend, poor recent recruitment, inability of the stock to rebuild, abrupt increase or decrease in stock abundance.
Environmental/ecosystem considerations–trends in environmental/ecosystem indicators, ecosystem model results, decreases in ecosystem productivity, decreases in prey abundance or availability, increases or increases in predator abundance or productivity.
Fisheries considerations–fishery CPUE is showing a contrasting pattern from the stock biomass trend, unusual spatial pattern of fishing, changes in the percent of TAC taken, changes in the duration of fishery openings.”
Assessment considerations
The EBS pollock assessment model has appeared to track the stock from year-to-year based on retrospective analysis in previous assessments. In 2022 the surveys showed an increase from the relatively low observation from 2021. This affected the retrospective analyses which last year indicated a tendency to over estimate the stock trend. This year the model tracks the available data reasonably well and fishery data confirm an apparently very strong 2018 year class. The trend towards fishing being below average in body weight so far has been limited, despite the observation that the condition (weight-given length) in 2023 seems to be the lowest on record. We therefore rated the assessment-related concern as Level 1, No Concern.
Population dynamics considerations
The age structure of EBS pollock has exhibited some peculiarities over time. On the positive side, some strong year-classes appear to have increased in abundance based on the bottom-trawl survey data (e.g., the 1992, 2012, 2013 and 2018 year classes). Conversely, the period from 2000–2007 had relatively poor year-class strengths which resulted in declines in stock below \({B_{msy}}\) and reduced TACs due to lower ABC values. Given new support for the strong year-class strength from 2018, it appears that the mean recruitment since 2000 has been nearly average but with greater variability than earlier years (Figure 66). The stock is estimated to be above \({B_{msy}}\) at present, and projections indicate a increases given recent catch levels. Recruitment in the near term is about average and highly uncertain. Additional age-specific aspects of the spawning population indicate that the stock has increased from a low diversity of ages (for both the population and the mean age of the spawning stock weighted by spawning output Figure 69). We therefore rated the population-dynamics concern as level 1, No Concern
Environmental/Ecosystem considerations
Summary for Environmental/Ecosystem considerations The following summarizes “Environmental/Ecosystem” considerations (see Section 15 for details):
September 2022 - August 2023 oceanographic conditions, based on sea surface and bottom temperatures, sea ice, and cold pool extent, were near respective time series averages.
The 2023 cold pool spatial extent was near its time series average (1982-2023).
2023 chlorophyll-a biomass was among the lowest in the time series
2023 coccolithophore index was among the highest ever observed in the timeseries
pH and Ωaragonite remain near threshold levels of biological significance
Zooplankton were dominated by small copepods. Large copepods and euphausiids had low lipid content; abundances increased to the north with hot spots around St. Lawrence Island
Age-0 pollock condition (length-weight residual, % lipid, and energy density residual) was below average in 2023 and continued decreasing trends
Juvenile pollock (100-250mm) condition was below average in the SEBS and NBS in 2021 and has decreased since 2021
Adult pollock (>250mm) condition was below average in the SEBS and has decreased since 2019, while condition was above average in the NBS and has increased since 2021
Trends in potential competitors are mixed over the shelf, with increases in jellyfish over the northern shelf, increases in herring biomass, mixed trends in salmon run strengths and juvenile salmon condition (negative over the southern shelf, positive over the northern shelf), and a continued decline in the biomass of the pelagic forager biomass.
In 2023, with an average cold pool extent, predation pressure from cannibalism may have been mitigated as adult pollock avoided the cold pool and their center of distribution shifted north and west.
Northern fur seal pup production at St. Paul Island in 2022 continued a declining trend since 1998.
Trends in other potential predators are mixed over the shelf, especially in the NBS, with increased jellyfish abundance and decreased chum salmon abundance.
Together, the most recent data available suggest an ecosystem risk of Level 2 – “Multiple indicators showing consistent adverse signals a) across the same trophic level as the stock, and/or b) up or down trophic levels (i.e., predators and prey of the stock).” Multiple indicators of primary and secondary productivity show adverse signals borne out in continued declining trends in juvenile and adult fish condition.
Fishery performance
As noted above, the 2023 fishery again experienced good nominal fishing rates that were improved over 2020 and 2021 conditions. The 2023 fishery seemed to have generally smaller-than-expected pollock in the fishery and there was indications that the fish were unusually skinny given their length. The fleet dispersion (the relative distance or spread of the fishery in space) as shown in the past has indicated that the seasonal dispersion levels increased slightly but was still relatively low (indicating relatively good fishing; Figure 10).
The CPUE of PSC species and other bycatch declined in 2023. Sablefish, herring and Chinook salmon bycatch rates (per hour of fishing) all decreased from 2021 (except for a slight increase in herring CPUE during the B season from low levels; Figure 3).
The way the ABC control rule interacts with actual fishing is worth considering. Specifically, given the 2 million t OY cap for all of groundfish, when the EBS pollock stock is above target levels, the fishing effort is lower (a lower F). As it approaches the target (\(B_{MSY}\)), it increases and then when it drops below, the fishing mortality rate is ratcheted downwards rapidly. This can be exacerbated when there are sudden unanticipated changes in survey estimates (like what was apparently the case in 2021 which caused the retrospective pattern to degrade). The mean weight-at-age for the 2021 B-season was near average, but in general, pollock were skinny given their length. However, concerns over the amount of 2-year old pollock in the 2020 fishery data has been ameliorated with continued positive signs of that year-class which is projected to be an abundant number of 5-year olds in 2023. For this reason, we conclude that that the fishery performance warrants a score of 1, No Concern.
These results are summarized as:
Having a score at level 2 suggests that adjustments to the ABC may be prudent. In the past, the SSC has considered factors similar to those presented above and selected an ABC based on Tier 3 estimates. Last year the SSC requested examining Tier 2 values as an alternative. Unlike Tier 3, using Tier 2 would have a constant buffer relative to the Tier 1 value (at about 11%). Setting the ABC to Tier 3 levels provides a very large buffer but one that could be warranted given that the impact on subsequent spawning biomass levels will be much more variable and have a high probability of dropping below the target stock size and result in much reduced future ABCs under the current FMP. It is worth noting that fishing at the full Tier 1 ABC would imply a more than doubling of effort and well exceed the 2 million t groundfish catch limit. Even fishing at a full Tier 3 ABC shows there is a relatively high probability of falling below \(B_{MSY}\) values or proxies thereof. Under our standard scenarios, Alternative 3 shows trajectories if fishing effort is held equal to the recent 5-year average. It is noteworthy that this provides stock sizes that have a good probability of being above targets and avoiding drastic reductions in yeild (lower overall variability in ABC/yields; Figure 68).
The SSC has requested “an explicit set of concerns that explain the ABC adjustment.” In response, we direct attention to the decision table () and the fact that the biological basis for the continued stock productivity has most to do with the OY constraint which has effectively maintained fishery production at around 1.3 million t since 1990. Demonstrations that would allow fishing to near \(F_{MSY}\) catch quantities would show that catch variability would be extremely high (and unrealistic given current capacity and OY limits for combined BSAI groundfish; Ianelli 2005). Furthermore, the frequency of being at much lower spawning stock sizes would be much higher, and would likely be riskier and fishing effort would need to be much higher. While the biological basis for ABC setting is founded in sound conservation of spawning biomass, the history of the current fishery productivity should inform desirable biomass. In only 6 of the 41 years since 1981 has the stock been below the \(B_{MSY}\) level (15% of the years). The mean spawning biomass over this period has averaged about 13% higher than the estimated \(B_{MSY}\). In terms of an actual “management target”, Punt et al. (2013) developed some robust estimators for \(B_{MEY}\) (Maximum Economic Yield) noting that a typical target would be 1.2\(\times B_{MSY}\) or about -6% lower than the mean value or a target female spawning biomass at 2.822 million t. It therefore seems worth considering developing an explicit harvest control rule that achieves the level of productivity observed over the past 30 years.
In recent years when the pollock biomass was estimated to be well above average, the catch was constrained by other factors. Specifically, the 2 million t BSAI groundfish catch limit and bycatch avoidance measures has an impact on the potential for large increases in catch. As the stock is presently estimated to be below \(B_{MSY}\), the maximum permissible ABC under the FMP can become the limiting factor for TAC specification. Unfortunately, this ABC can ratchet down quickly because as the stock declines further below this target stock size, the ABC fishing mortality rate is adjusted downwards nearly proportionately. This part of the FMP control rule can create high variability in the TAC. Less variability in the catch, accordingly, would also result in less spawning stock variability and reduce risks to the fishery should the period of poor recruitment continue.
To more fully evaluate these considerations performance indicators as modified from Ianelli et al. (2012) were developed to evaluate some near-term risks given alternative 2025 catch values. These indicators and rationale for including them are summarized in ). Model 23 (the “base”) results for these indicators are provided in . Each column of this table uses a fixed 2025 catch and assumes the same effort for the four additional projection years (2026–2029). Given this specification, there is a low probability that any of the catches shown in the first row would exceed the \(F_{MSY}\) level. Also, in the near term it appears unlikely that the spawning stock will be below \(B_{MSY}\) (rows 3 and 4). Relative to the historical mean spawning biomass, by 2025 it is more likely than not that the spawning biomass will be lower than the historical mean (fifth row). The range of catches examined have relatively small or no impact on the age diversity indicators. The table indicates that for the 2024 catch to equal the 2023 value, about the same level of fishing effort would be required. In terms of catch advice, the results presented in the decision table indicates that catches above 1.3 million t will very likely result in 2026 spawning stock estimates being below the long term mean (but above \(B_{MSY}\)).
In the past, another approach/rationale for stabilizing effort by setting the fishing mortality equal to the current year. Doing so this year suggests setting the fishing mortality to 2024 levels results in a catch of 1,150,000 t. Given the revisions to last year’s model results and the positive increases in stock size, maintaining a constant fishing mortality rate seems unnecessary at this time.
Additional ecosystem considerations
In general, a number of key issues for ecosystem conservation and management can be highlighted. These include:
Preventing overfishing;
Avoiding habitat degradation;
Minimizing incidental bycatch;
Monitoring bycatch and the level of discards; and
Considering multi-species trophic interactions relative to harvest policies.
For the case of pollock in the Eastern Bering Sea, the NPFMC and NMFS continue to manage the fishery on the basis of these issues in addition to the single- species harvest approach (Hollowed et al. 2011). The prevention of overfishing is clearly set out as the main guideline for management. Habitat degradation has been minimized in the pollock fishery by converting the industry to pelagic-gear only. Bycatch in the pollock fleet is closely monitored by the NMFS observer program and managed on that basis. Discard rates of many species have been reduced in this fishery and efforts to minimize bycatch continue.
In comparisons of the Western Bering Sea (WBS) with the Eastern Bering Sea using mass-balance food-web models based on 1980–85 summer diet data, Aydin et al. (2002) found that the production in these two systems is quite different. On a per-unit-area measure, the western Bering Sea has higher productivity than the EBS. Also, the pathways of this productivity are different with much of the energy flowing through epifaunal species (e.g., sea urchins and brittlestars) in the WBS whereas for the EBS, crab and flatfish species play a similar role. In both regions, the keystone species in 1980–85 were pollock and Pacific cod. This study showed that the food web estimated for the EBS ecosystem appears to be relatively mature due to the large number of interconnections among species. In a more recent study based on 1990–93 diet data, pollock remain in a central role in the ecosystem. The diet of pollock is similar between adults and juveniles with the exception that adults become more piscivorous (with consumption of pollock by adult pollock representing their third largest prey item).
Regarding specific small-scale ecosystems of the EBS, Ciannelli et al. (2004a, 2004b) presented an application of an ecosystem model scaled to data available around the Pribilof Islands region. They applied bioenergetics and foraging theory to characterize the spatial extent of this ecosystem. They compared energy balance, from a food web model relevant to the foraging range of northern fur seals and found that a range of 100 nautical mile radius encloses the area of highest energy balance representing about 50% of the observed foraging range for lactating fur seals. This has led to a hypothesis that fur seals depend on areas outside the energetic balance region. This study develops a method for evaluating the shape and extent of a key ecosystem in the EBS (i.e., the Pribilof Islands). Furthermore, the overlap of the pollock fishery and northern fur seal foraging habitat has been identified (see Sterling and Ream (2004), Zeppelin and Ream (2006)).
A brief summary of these two perspectives (ecosystem effects on pollock stock and pollock fishery effects on ecosystem) is given in (). Unlike the food-web models discussed above, examining predators and prey in isolation may overly simplify relationships. This table serves to highlight the main connections and the status of our understanding or lack thereof.
Ecosystem effects on the EBS pollock stock
The pollock stock trends appear to be responding to ecosystem conditions in the EBS. The conditions on the shelf during 2008 apparently affected age-0 northern rock sole due to cold conditions and apparently unfavorable currents that retain them into the over- summer nursery areas (Cooper et al. (2014)). It may be that such conditions favor pollock recruitment. Hollowed et al. (2012) provided an extensive review of habitat and density for age-0 and age-1 pollock based on survey data. They noted that during cold years, age-0 pollock were distributed primarily in the outer domain in waters greater than 1\(^\circ\)C and during warm years, age-0 pollock were distributed mostly in the middle domain. This temperature relationship, along with interactions with available food in early-life stages, appears to have important implications for pollock recruitment success (Coyle et al. (2011)).
A separate section presented again this year updates a multispecies model with more recent data and is presented as a supplement to the BSAI SAFE report. This approach incorporates a number of simplifications for the individual species data and fisheries processes (e.g., constant fishery selectivity and the use of design-based survey indices for biomass). However, that model mimics the biomass levels and trends with the single species reasonably well. It also allows specific questions to be addressed regarding pollock TACs. For example, since predation (and cannibalism) is explicitly modeled, the impact of relative stock sizes on subsequent recruitment to the fishery can be now be directly estimated and evaluated (in the model presented here, cannibalism is explicitly accounted for in the assumed Ricker stock-recruit relationship).
EBS pollock fishery effects on the ecosystem.
Since the pollock fishery is primarily pelagic in nature, the bycatch of non- target species is small relative to the magnitude of the fishery (). Jellyfish represent the largest component of the bycatch of non-target species and had averaged around 5–6 kt per year but more than doubled in 2014, then dropping again in 2015. The 2018 value was high, dropped and then was again high in 2021. The data on non-target species shows a high degree of inter-annual variability, which reflects the spatial variability of the fishery and high observation error. This variability may reduce the ability to detect significant trends for bycatch species.
The catch of other target species in the pollock fishery (defined as any trawl set where the catch represents more than 80% of the catch) represents about 1% of the total pollock catch. Incidental catch of Pacific cod has varied but after a period of low catch levels it increased to over 9,000 t in 2020 and 2021 but in 2022 was under 4 thousand t (). There has been a marked increase in the incidental catch of Pacific ocean perch in the since 2014 with a peak just under 8 thousand t in 2019. The incidental catch of sablefish peaked in 2020 at about 3.5 thousand t but was less that 300 t in 2022. The incidental catch of pollock in other target fisheries is more than double the bycatch of target species in the pollock fishery with the largest pollock catches in the yellofin sole and Pacific cod fisheries ().
The number of non-Chinook salmon (nearly all made up of chum salmon) taken incidentally varies considerably over time. The bycatch increased since 2014 with the 2017 number in excess of 465 thousand fish, the third highest non-Chinook salmon bycatch that’s been observed since 1991. Since then, 7 of the top 10 highest bycatch years have occurred with nearly 550 thousand taken in 2021 (). Chinook salmon bycatch has varied (42% CV since 2011) and averaged just under 19 thousand fish from 2011-2023 (). After a recent high bycatch of over 32,000 fish in 2020, the 2022 and 2023 bycatch was 6,415 and 11,750 Chinook salmon, respectively. J. N. Ianelli and Stram (2014) provided estimates of the bycatch impact on Chinook salmon runs to the coastal west Alaska region and found that the peak bycatch levels exceeded 7% of the total run return. Since 2011, the impact has been estimated to be below 2%. Updated estimates given new genetic information and these levels of PSC as provided to the Council continue to suggest that the impact is low.
Data gaps and research priorities
The available data for EBS pollock are extensive yet many processes behind the observed patterns continue to be poorly understood. The recent patterns of abundance observed in the northern Bering Sea provide an example. As such, we recommend the following research priorities:
Support developing a team of analysts to evaluate all aspects of the current model against alternatives (e.g., Rceattle, WHAM, Stock Synthesis, etc.). This work has progressed and presently, developments on the Gulf of Alaska assessment adopting WHAM appears promising.
Continue to investigate using spatial processes for estimation purposes (e.g., combining acoustic and bottom trawl survey data). The application of the geostatistical methods seems like a reasonable approach to statistically model disparate data sources for generating better abundance indices. Also, examine the potential to use pelagic samples from the BASIS survey to inform recruitment and subsequent spatial patterns. Work on developing data from the BASIS survey to inform recruitment has not been pursued. The work to refine the AVO data has helped with information used in this assessment.
Develop methods to use spatio-temporal models to estimate composition information (specifically, weight-at-age in the survey). Two papers, (Indivero et al. (2023)) and (Cheng et al. (2023)) have been published targeting this type of activity. The former is presently used in this assessment while the latter has yet to be applied.
Study the relationship between climate and recruitment and trophic interactions of pollock within the ecosystem. This would be useful for improving ways to evaluate the current and alternative fishery management systems. In particular, a careful re-evaluation of the current FMP harvest control rule should be undertaken. As part of the ACLIM program, progress is being made. However, a full evaluation of the FMP control rules is pending.
Apply new technologies (e.g., bottom-moored echosounders) to evaluate pollock movement between regions and supplement this work with analytical approaches. The data have been processed completely and a manuscript is being submitted. Next steps is to use this information to develop scenarios for flux over the maritime boundary and evaluate relative fishing effort impacts on either side.
Expand genetic sample collections for pollock (and process available samples) and apply high resolution genetic tools for stock structure analyses. Additional analyses have been completed and are expected to lead to a publication in 2024.
Acknowledgments
We thank the scientifically trained observers and the staff of the Fisheries and Monitoring Division at AFSC for their hard work. The diligence of survey staff who contribute immensely in collecting samples, especially given these times is exceptional. The AFSC age-and-growth department is thanked for their continued excellence in promptly processing the samples used in this assessment. We thank the many colleagues who provided edits and suggestions to improve this document, in particular, the timely review done by Dr. Melissa Haltuch.
References
Tables
% latex table generated in R 4.4.1 by xtable 1.8-4 package % Wed Oct 23 10:37:19 2024 \scalebox{0.9}{ \scalebox{0.9}{Figures
Analytic approach
General model structure
A statistical age-structured assessment model conceptually outlined in Fournier and Archibald (1982) and like Methot’s (1990) stock synthesis model was applied over the period 1964–2023. A technical description is presented in the “EBS Pollock Model Description” appendix. The analysis was first introduced in the 1996 SAFE report and compared to the cohort analyses that had been used previously and was later documented in Ianelli and Fournier (1998). The model was written in ADMB—a library for non-linear estimation and statistical applications (Fournier et al. 2012). The data updated from last year’s analyses include:
The 2022 fishery age composition data were added
The catch biomass estimates were updated through to the current year
The 2023 bottom-trawl survey index, weight, and age composition data were revised and added
The 2023 acoustic-trawl survey index, weight, and preliminary age composition data were added
The AVO backscatter data collected opportunistically from the 2022 bottom trawl survey and post processed into the AVO backscatter index were included.
A simplified version of the assessment (with mainly the same data and likelihood-fitting method) is included as a supplemental multi-species assessment model. As presented since 2016, it allows for trophic interactions among key prey and predator species and for pollock, and it can be used to evaluate age and time-varying natural mortality estimates in addition to alternative catch scenarios and management targets (see this volume: EBS multi-species model).
Description of alternative models
In the 2019 assessment, the spatio-temporal model fit to BTS CPUE data including stations from the NBS was expanded using the VAST methods detailed in Thorson (2018). This data treatment was included as a model alternative and adopted for ABC/OFL specifications by the SSC in 2020 along with other modifications including a spatio-temporal treatment of the age composition data. This year, we used the same model configuration and simply examined the influence of additional data that became available this year. For projections we added the ability to test alternative Tier scenarios. The current base model is Model 20.0a, which was adopted last year by the SSC, and which differed from the previous base model (Model 16.2) in that it included the 2020 USV acoustic biomass estimate as an extension of the standard AT survey biomass time series and excluded the 1978 year class from the estimation of the stock-recruitment relationship. We examined the following alternative models:
In an effort to test different stock assessment software, we adopted the pollock data and used some of the model software that was simulation tested in Li et al. (2021). While preliminary, the results were consistent with the bespoke model used here. However, the impact of missing features in the more generalized models tested in this paper requires more investigation. Specifically, the bespoke model used here includes an informed fixed-effects model for projecting weight-at-age and uses a covariance matrix for index time series that is unavailable in the models tested in Li et al. (2021).
Input sample size
Sample sizes for age-composition data were re-evaluated in 2016 against the trade-off with flexibility in time and age varying selectivity. In subsequent assessment years the values have changed significantly from the 4-periods of fishery data from which these weights were applied and calculated. Principally, this work resulted in tuning the recent era (1991-present year) to an average sample sizes of 350 for the fishery and then using estimated values for the period 1978-1990 and as earlier (Table \(\ref{tab:input_n}\)). We assumed average values of 100 and 50 for the BTS and ATS data, respectively with inter-annual variability reflecting the variability in the number of hauls sampled for ages. This year we re-evaluated one-step tuning as a sensitivity following Francis 2011 (equation TA1.8, hereafter referred to as Francis weights).
Recent work has shown ways to improve estimation schemes that deal with the interaction between flexibility in fishery selectivity and statistical properties of composition data sample size. Specifically, the Dirichlet-multinomial using either Laplace approximation (Thorson et al., 2015) or adnuts (Monnahan and Kristensen, 2018) should be implemented (e.g., as shown by Xu et al., 2020). We hope to evaluate these and alternatives in the coming year.
Parameters estimated outside of the assessment model
Natural mortality and maturity at age
The baseline model specification has been to use constant natural mortality rates at age (M=0.9, 0.45, and 0.3 for ages 1, 2, and 3+ respectively (Wespestad and Terry 1984). When predation was explicitly considered estimates tended to be higher and more variable (Holsman et al. this volume; Holsman et al. 2015; Livingston and Methot 1998; Hollowed et al. 2000). Clark (1999) found that specifying a conservative (lower) natural mortality rate may be advisable when natural mortality rates are uncertain. More recent studies confirm this (e.g., Johnson et al. 2015).
In the supplemental multi-species assessment model alternative values of age and time-varying natural mortality are presented. As in past years the estimates indicate higher values than used here. In the 2018 assessment we evaluated natural mortality, and it was noted that the survey age compositions favored lower values of M while the fishery age composition favored higher values. This is consistent with the patterns seen in the BTS survey data as they show increased abundances of “fully selected” cohorts. Hence, given the model specification (asymptotic selectivity for the BTS age composition data), lower natural mortality rates would be consistent with those data. Given these trade-offs, structural model assumptions were held to be the same as previous years for consistency (i.e., the mortality schedule presented below).
Maturity-at-age values used for the EBS pollock assessment were originally based on Smith (1981) and were later reevaluated (e.g., Stahl 2004; Stahl and Kruse 2008a; and Ianelli et al. 2005). These studies found inter-annual variability but general consistency with the original schedule of proportion mature at age.
Based on results from a distinct (apparently) but adjacent stock (Bogoslof assessment, this volume) where fishing has been curtailed since 1992 and spawning surveys have taken place with regularity since then (and included age data) we evaluate as a sensitivity estimated natural mortality for pollock age 3-yrs and older. For the “base” model (model 2020) we continue to use assumed natural mortality-at-age and maturity-at-age (for all models; Smith 1981) as in previous assessments:
Length and weight-at-age
Age determination methods have been validated for pollock (Kimura et al. 1992; Kimura et al. 2006, and Kastelle and Kimura 2006). EBS pollock size-at-age show important differences in growth with differences by area, year, and year class. Pollock in the northwest area are typically smaller at age than pollock in the southeast area. The differences in average weight-at-age are taken into account by stratifying estimates of catch-at-age by year, area, season, and weighting estimates proportional to catch.
The assessment model for EBS pollock accounts for numbers of individuals in the population. As noted above, management recommendations are based on allowable catch levels expressed as tons of fish. While estimates of pollock catch-at-age are based on large data sets, the data are only available up until the most recent completed calendar year of fishing (e.g., 2021 for this year). Consequently, estimates of weight-at-age in the current year are required to map total catch biomass (typically equal to the quota) to numbers of fish caught (in the current year). Therefore, if there are errors (or poorly accounted uncertainty) in the current and future mean weight-at-age, this can translate directly into errors between the expected fishing mortality and what mortality occurs. For example, if the mean weight-at-age is biased high, then an ABC (and OFL) value will result in greater numbers of fish being caught (and fishing mortality being higher due to more fish fitting within the ABC).
As in previous assessments, we explored patterns in size-at-age and fish condition. Using the NMFS fishery observer data on weight given length we:
extracted all data where non-zero measurements of pollock length and weight were available between the lengths of 35 and 60 cm for the EBS region
computed the mean value of body mass (weight) for each cm length bin over all areas and time
divided each weight measurement by that mean cm-specific value (the “standardization” step)
plotted these standardized values by different areas, years, months etc. to evaluate condition differences (pooling over ages is effective as there were no size-specific biases apparent)
In the first instance, the overarching seasonal pattern in body mass relative to the mean shows that as the winter progresses prior to peak spawning, pollock are generally skinnier than average whereas in July, the median is about average (Fig. \(\ref{fig:fsh_lw_month}\)). As the summer/fall progresses, fish were at their heaviest given length (Fig. \(\ref{fig:fsh_lw_month}\)). This is also apparent when the data are aggregated by A- and B-seasons (and by east and west of 170\(^\circ\)W; referred to as SE and NW respectively) when plotted over time (Fig. \(\ref{fig:fsh_lw_anom_str_yr_box}\), where stratum 1 = A season, stratum 2 = B season SE, and stratum 3 = B season NW). Combining across seasons, the fishery data shows that recent years were below average weight given length (Fig. \(\ref{fig:fsh_lw_anom_yr_box}\); note that the anomalies are based on the period 1991-2022).
Examining the weight-at-age, there are also patterns of variability that vary due to environmental conditions in addition to spatial and temporal patterns of the fishery. Based on the bootstrap distributions and large sample sizes, the within-year sampling variability for pollock is small. However, the between-year variability in mean weights-at-age is relatively high (Table \(\ref{tab:wtage}\)). The coefficients of variation between years are on the order of 6% to 9% (for the ages that are targeted) whereas the sampling variability is generally around 1% or 2%. The approach to account for the identified mean weight-at-age having clear year and cohort effects was continued (e.g., Fig. \(\ref{fig:fsh_wtage_comb}\)). Details were provided in appendix 1A of Ianelli et al. (2016). The results from this method showed the relative variability between years and cohorts and provide estimates for 2023–2025 (Table \(\ref{tab:wtage}\)). How these fishery weights-at-age estimates can be supplemented using survey weights-at-age is further illustrated in Fig. \(\ref{fig:fish_wtage_data_pred}\).
In the 2020 and 2021 fishery, the average weight-at-age for ages 6-8 (the 2012-2014 year classes) was below the time series average. These cohorts have fluctuated around their means in recent years (Fig. \(\ref{fig:fsh_wtage_comb}\)). To examine this more closely, we split the bootstrap results into area-season strata and were able to get an overall picture of the pattern by strata (Fig. \(\ref{fig:fsh_wtage_strata}\) and Fig. \(\ref{fig:fsh_wtage_strata_yr}\)). This showed that the mean weight-at-age is higher in the the B-season in the area east of 170\(^\circ\)W compared to the A-season and B-season in the area west of 170\(^\circ\)W.
Parameters estimated within the assessment model
Fishing mortality is parameterized to be semi-separable with year and age (selectivity) components. The age component is allowed to vary over time; changes are allowed in each year. The mean value of the age component is constrained to equal one and the last 5 age groups (ages 11–15) are specified to be equal. This latter specification feature is intended to reduce the number of parameters while acknowledging that pollock in this age-range are likely to exhibit similar life-history characteristics (i.e., unlikely to change their relative availability to the fishery with age). The annual components of fishing mortality result in 60 parameters and the age-time selectivity schedule forms a 10x60 matrix of 600 parameters bringing the total fishing mortality parameters to 660. The rationale for including time- varying selectivity has recently been supported as a means to improve retrospective patterns (Szuwalski et al. 2017) and as best practice (Martell and Stewart, 2013).
For surveys and indices, the treatment of the catchability coefficient, and interactions with age-specific selectivity require consideration. For the BTS index, selectivity-at-age is estimated with a logistic curve in which year specific deviations in the parameters is allowed. Such time-varying survey selectivity is estimated to account for changes in the availability of pollock to the survey gear and is constrained by pre-specified variance terms. Presently, these variance terms have been set based on balancing input data-based variances and are somewhat subjective. For the AT survey, which originally began in 1979 (the current series including data down to 0.5 m from bottom begins in 1994), optional parameters to allow for age and time-varying patterns exist but for this assessment and other recent assessments, ATS selectivity is constant over time. Overall, four catchability coefficients were estimated: one each for the early fishery catch-per-unit effort (CPUE) data (from Low and Ikeda, 1980), the VAST combined bottom trawl survey index, the AT survey data, and the AVO data. An uninformative prior distribution is used for all of the indices. The selectivity parameters for the 2 main indices (BTS and ATS) total 135 (the CPUE and AVO data mirror the fishery and AT survey selectivities, respectively).
Additional fishing mortality rates used for recommending harvest levels are estimated conditionally on other outputs from the model. For example, the values corresponding to the \(F_{40\%}\) \(F_{35\%}\) and \(F_{MSY}\) harvest rates are found by satisfying the constraint that, given age-specific population parameters (e.g., selectivity, maturity, mortality, weight-at-age), unique values exist that correspond to these fishing mortality rates. The likelihood components that are used to fit the model can be categorized as:
- Total catch biomass (log-normal, \(\sigma=0.05\))
- Log-normal indices of pollock biomass; bottom trawl surveys assume annual estimates of sampling error, as represented in Fig. \(\ref{fig:bts_biom}\) along with the covariance matrices (for the density-dependent and VAST index series); for the AT index the annual errors were specified to have a mean CV of 0.20; while for the AVO data, a value relative to the AT index was estimated and scaled to have a mean CV of 0.3).
- Fishery and survey proportions-at-age estimates (multinomial with effective sample sizes presented Table \(\ref{tab:input_n}\)).
- Age 1 index from the AT survey (CV set equal to 30% as in prior assessments).
- Selectivity constraints: penalties/priors on age-age variability, time changes, and decreasing (with age) patterns.
- Stock-recruitment: penalties/priors involved with fitting a stochastic stock-recruitment relationship within the integrated model.
- “Fixed effects” terms accounting for cohort and year sources of variability in fishery mean weights-at-age estimated based on available data from 1991-2022 from the fishery (and 1982-2023 for the bottom-trawl survey data) and externally estimated variance terms as described in Appendix 1A of Ianelli et al. (2016; see Fig. \(\ref{fig:fish_wtage_data_pred}\)).
Work evaluating temperature and predation-dependent effects on the stock- recruitment estimates continues (Spencer et al. 2016). This approach modified the estimation of the stock-recruitment relationship by including the effect of temperature and predation mortality. A relationship between recruitment residuals and temperature was noted(similar to that found in Mueter et al., 2011 and subsequently noted in Thorson et al., 2020a) and lower pollock recruitment during warmer conditions might be expected. Similar results relating summer temperature conditions to subsequent pollock recruitment for recent years were also found by Yasumiishi et al. (2015) where research suggests that summer warmth is associated with earlier diapause of copepods (Thorson et al., 2020b), such that a fall (but not spring) survey of copepod densities is also associated with cold conditions and elevated recruitment (Eisner et al., 2020).
Appendix 1. Risk Table information for Environmental/Ecosystem considerations
Provided by: Elizabeth Siddon, NOAA/AFSC
Environmental processes
The recent warm stanza in the eastern Bering Sea (EBS) persisted from approximately 2014 through 2021. Since 2021, the EBS has experienced cooler oceanographic conditions with the most recent year’s metrics of sea surface and bottom temperatures, sea ice, and cold pool extent being near their respective time series averages. Regional sea surface and bottom temperature trends were largely at or near the long-term average in 2023. Exceptions include (i) slightly warmer than average sea surface temperature (SST) over the outer domain (southern and northern shelf) and over the southern middle domain from approximately December 2022 through April 2023 and (ii) slightly cooler than average bottom temperature over the outer domain of the southern shelf from August 2022 through August 2023. Due to these exceptions, the outer domain may have been a more strongly vertically stratified system during late winter/early spring 2022-2023. During the standard bottom trawl survey in summer 2023, bottom temperatures were slightly cooler than the time series average with the coldest bottom temperatures in the southern inner domain since 2013. Marine heatwaves based on SSTs have been brief and infrequent in the EBS since January 2021 (Hennon et al., 2023).
Age-0 fish experiencing warm temperatures during late summer followed by relatively cooler temperatures in spring of age-1 are thought to have below average survival. Based on this Temperature Change index, the 2022 year class is predicted to have average recruitment to age-4 in 2026 (Yasumiishi, 2023).
Sea ice metrics, such as early season (Oct. - Dec.) ice extent, annual ice extent, and sea ice thickness were all near their respective time series averages. The 2023 cold pool extent was also near its historical average. Broad-scale climate indices, like the North Pacific Index, reflected a transition from La Niña conditions to developing El Niño conditions in the tropic Pacific; the National Multi-Model Ensemble predicts SST anomalies of +0.5-1°C over the SEBS shelf through May 2024 (Hennon et al., 2023).
The center of gravity estimate for pollock has moved steadily north since approximately 2000, but shifted fairly south in 2022, and then north again in 2023 (though not to the most northward position). The stock center of gravity also moved east from 2010 through 2017, then shifted west to approximately its time series average in 2023. The area occupied by the stock steadily expanded since 2010, with the exception of 2018, and has since decreased since 2019 but remains above the time series average in 2023 (Figure 19).
Prey
Chlorophyll-a biomass was among the lowest in every sub-region over the southern and northern shelf and slope for 2023 (Nielsen et al., 2023). Additionally, in 2023, the coccolithophore index for both the inner and middle shelf was among the highest ever observed in the timeseries. The striking milky aquamarine color of the water during a coccolithophore bloom can reduce foraging success for visual predators, such as fish and surface-feeding seabirds (Nielsen and Eisner, 2023). Modeled bottom pH and Ωarag (important for pteropod shell formation) both improved this year compared to lowest to near-lowest values for the model hindcast in 2022, though both remain near threshold levels of biological significance (Pilcher et al., 2023).
Small copepods form the prey base for larval to early juvenile pollock during spring. Late juvenile pollock feed on a variety of planktonic crustaceans, including large calanoid copepods and euphausiids. The Rapid Zooplankton Assessment in the southeastern Bering Sea in spring noted a moderate abundance of small copepods, but low abundance and low lipid content of large copepods and euphausiids. In fall, the moderate abundance of small copepods continued, and while the abundance of large copepods and euphausiids remained low, abundances increased from south to north. In the northern Bering Sea in fall, small copepods were ubiquitous and increased in abundance from south to north, while hot spots of large copepods and euphausiids were observed around St. Lawrence Island (Kimmel et al. 2023).
A significant relationship exists between the abundance of large, lipid-rich copepods and the recruitment and survival of juvenile pollock to age-3 (Eisner et al., 2020). Low availability of large copepod prey in 2020 and 2022 may result in reduced overwinter survival and recruitment to age-3 (in 2023 and 2025) (Yasumiishi et al., 2023b).
Age-0 fish condition (samples collected during southern and northern Bering Sea surface trawl surveys), measured by length-weight residuals, % lipid content, and energy density residuals, were all below their respective time series averages in 2023. Length-weight and energy density residuals show decreasing trends since 2021 while the mean % lipid has been below average (time series 2002-2023) since the beginning of the recent warm stanza in 2014 (Page et al., 2023). Juvenile (100-250 mm) fish condition in the southern Bering Sea decreased since 2021 while adult fish condition has decreased since 2019 with 2023 being the second lowest in the time series (1997-2023). Juvenile fish condition in the northern Bering Sea has decreased since 2021 while adult fish condition has increased since 2021 with 2023 being the highest in the time series (2010, 2017, 2019, 2021-2023) (Prohaska and Rohan, 2023).
Competitors
Jellyfish feed primarily on zooplankton and small fish, and therefore may compete for prey resources for both juvenile and adult life stages of pollock. Jellyfish abundance over the southeastern Bering Sea shelf was average over the time series (1982-2023; Buser, 2023), whereas abundance increased over the northern Bering Sea shelf (Buser, 2023; Yasumiishi et al., 2022c). Togiak herring biomass has been increasing in recent years as a result of strong 2016 and 2017 year classes that have also contributed to high Prohibited Species Catch in the EBS pollock fishery (Joy et al., 2023).
Yukon and Kuskokwim River salmon runs have experienced precipitous declines in recent years (Whitworth et al., 2023), leading to potential reduced competition for prey resources where these stocks overlap, though slight increases were observed in juvenile Chinook and chum salmon indices in 2023 (Murphy et al., 2023). In 2023, returns of Bristol Bay sockeye salmon remained exceptionally high relative to the long-term average, but lower than the recent record high numbers observed in 2021 and 2022 (Cunningham and Vega, 2023). Juvenile sockeye salmon feed on zooplankton (competitors with age-0 pollock) and age-0 pollock (competitors with adult pollock) in warm years; adults feed on zooplankton and krill. Juvenile salmon condition in 2022, measured by energy density anomalies, was negative for all species (except neutral for Chinook salmon) in the southeastern Bering Sea and positive for all species (except neutral for sockeye salmon) in the northern Bering Sea (Fergusson et al., 2023).
The biomass of pelagic foragers measured during the standard EBS bottom trawl survey decreased 34% from 2022 to 2023 and is below the long-term mean (1982-2023) and is now at its third lowest value. The guild is largely driven by walleye pollock that decreased 25% from 2022. Pacific herring decreased 75% from a time series high in 2022, but remain above their long-term mean (Siddon, 2023). The impacts of recent large year classes of sablefish to the EBS ecosystem (as prey, predators, and competitors) remain largely unknown at this time, but may compete with pollock for prey resources as juveniles.
Predators
Pollock are cannibalistic and rates of cannibalism might be expected to increase as the biomass of older, larger fish increases (i.e., the aging of the large 2018 year class). In 2023, with an average cold pool extent over the shelf, predation pressure from cannibalism may have been mitigated by this thermal barrier as adult pollock tend to avoid the cold bottom waters. However, the biomass of pelagic foragers, including adult pollock, dropped to their third lowest value over the time series in 2023 (Siddon, 2023). Fur seal consumption of adult pollock generally increases in years when juvenile pollock are less abundant (Kuhn et al., 2019). However, Northern fur seal pup production at St. Paul Island in 2022 continued a declining trend since 1998 that may be partially attributed to low pup growth rates. Other potential predators of juvenile pollock include jellyfish and chum salmon. Jellyfish abundance over the southeastern Bering Sea shelf was average over the time series (1982-2023; Buser, 2023), whereas abundance increased over the northern Bering Sea shelf (Buser, 2023; Yasumiishi et al., 2022c). Chum salmon abundance has been declining in the Yukon and Kuskokwim Rivers since 2017 (Whitworth et al., 2023), a trend also reflected in the commercial harvest data (Whitehouse, 2023).
Additional references
Buser, T. 2023. Eastern and Northern Bering Sea – Jellyfishes. In: E.C. Siddon, 2023. Ecosystem Status Report 2023: Eastern Bering Sea, Stock Assessment and Fishery Evaluation Report, North Pacific Fishery Management Council, 1007 West Third, Suite 400, Anchorage, Alaska 99501
Cunningham, C.J. and S. Vega. 2023. Temporal Trend in the Annual Inshore Run Size of BristolBay Sockeye Salmon (Oncorhynchus nerka). In: E.C. Siddon, 2023. Ecosystem Status Report 2023: Eastern Bering Sea, Stock Assessment and Fishery Evaluation Report, North Pacific Fishery Management Council, 1007 West Third, Suite 400, Anchorage, Alaska 99501
Eisner, L.B., E.M. Yasumiishi, A.G. Andrews III, and C.A. O’Leary. 2020. Large copepods as leading indicators of walleye pollock recruitment in the southeastern Bering Sea: Sample-Based and spatio-temporal model (VAST) results. Fisheries Research 232:105720. Fergusson, E., R. Suryan, T. Miller, J. Murphy, and A. Andrews. 2023. Juvenile Salmon Condition Trends in the Eastern Bering Sea. In: E.C. Siddon, 2023. Ecosystem Status Report 2023: Eastern Bering Sea, Stock Assessment and Fishery Evaluation Report, North Pacific Fishery Management Council, 1007 West Third, Suite 400, Anchorage, Alaska 99501
Hennon, T., L. Barnett, N. Bond, M. Callahan, L. Divine, K. Kearney, E. Lemagie, A. Lestenkof, J. Overland, S. Rohan, K. Siwicke, R. Thoman, and M. Wang. 2023. Physical Environment Synthesis. In: E.C. Siddon, 2023. Ecosystem Status Report 2023: Eastern Bering Sea, Stock Assessment and Fishery Evaluation Report, North Pacific Fishery Management Council, 1007 West Third, Suite 400, Anchorage, Alaska 99501
Joy, P., S. Dressel, S. Miller, C. Brown, and J. Erickson. 2023. Togiak Herring Population Trends. In: E.C. Siddon, 2023. Ecosystem Status Report 2023: Eastern Bering Sea, Stock Assessment and Fishery Evaluation Report, North Pacific Fishery Management Council, 1007 West Third, Suite 400, Anchorage, Alaska 99501
Kimmel, D., D. Cooper, B. Cormack, C. Harpold, J. Murphy, M. Paquin, C. Pinger, B. Snyder, and R. Suryan. 2023. Current and Historical Trends for Zooplankton in theBering Sea. In: E.C. Siddon, 2023. Ecosystem Status Report 2023: Eastern Bering Sea, Stock Assessment and Fishery Evaluation Report, North Pacific Fishery Management Council, 1007 West Third, Suite 400, Anchorage, Alaska 99501
Kuhn, C., Sterling, J., and McHuron, E. 2019. Contrasting Trends in Northern Fur Seal Foraging Effort Between St. Paul and Bogoslof Islands: 2019 Preliminary Results. In: Siddon, E., and Zador, S., 2019. Ecosystem Status Report 2019: Eastern Bering Sea, Stock Assessment and Fishery Evaluation Report, North Pacific Fishery Management Council, 605 W 4th Ave, Suite 306, Anchorage, AK 99501.
Murphy, J., S. Garcia, A. Dimond, D. Cooper, E. Lee, and K. Howard. 2023. Northern Bering Sea Juvenile Salmon Abundance Indices. In: E.C. Siddon, 2023. Ecosystem Status Report 2023: Eastern Bering Sea, Stock Assessment and Fishery Evaluation Report, North Pacific Fishery Management Council, 1007 West Third, Suite 400, Anchorage, Alaska 99501
Nielsen, J. and L. Eisner. 2023. Coccolithophores in the Bering Sea. In: E.C. Siddon, 2023. Ecosystem Status Report 2023: Eastern Bering Sea, Stock Assessment and Fishery Evaluation Report, North Pacific Fishery Management Council, 1007 West Third, Suite 400, Anchorage, Alaska 99501
Nielsen, J.M., M.W. Callahan, L. Eisner, J. Watson, J.C. Gann, C.W. Mordy, S.W. Bell, and P. Stabeno. 2023. Spring Satellite Chlorophyll-a Concentrations in the Eastern Bering Sea. In: E.C. Siddon, 2023. Ecosystem Status Report 2023: Eastern Bering Sea, Stock Assessment and Fishery Evaluation Report, North Pacific Fishery Management Council, 1007 West Third, Suite 400, Anchorage, Alaska 99501
Page, J., J. Maselko, R. Suryan, T. Miller, E. Siddon, C. Pinger, E. Fergusson, and B. Cormack. 2023. Fall Condition of Young-Of-The-Year Walleye Pollock in the Southeastern and Northern Bering Sea, 2002–2023. In: E.C. Siddon, 2023. Ecosystem Status Report 2023: Eastern Bering Sea, Stock Assessment and Fishery Evaluation Report, North Pacific Fishery Management Council, 1007 West Third, Suite 400, Anchorage, Alaska 99501
Pilcher, D. J. Cross, N. Monacci, E. Kennedy, E. Siddon, and W.C. Long. 2023. Ocean Acidification. In: E.C. Siddon, 2023. Ecosystem Status Report 2023: Eastern Bering Sea, Stock Assessment and Fishery Evaluation Report, North Pacific Fishery Management Council, 1007 West Third, Suite 400, Anchorage, Alaska 99501
Prohaska, B. and S. Rohan. 2023. Eastern and Northern Bering Sea Groundfish Condition. In: E.C. Siddon, 2023. Ecosystem Status Report 2023: Eastern Bering Sea, Stock Assessment and Fishery Evaluation Report, North Pacific Fishery Management Council, 1007 West Third, Suite 400, Anchorage, Alaska 99501
Siddon, E.C., 2023. Eastern Bering Sea 2023 Report Card. In: E.C. Siddon, 2023. Ecosystem Status Report 2023: Eastern Bering Sea, Stock Assessment and Fishery Evaluation Report, North Pacific Fishery Management Council, 1007 West Third, Suite 400, Anchorage, Alaska 99501
Whitehouse, G.A. 2023. Trends in Alaska Commercial Salmon Catch – Bering Sea. In: Siddon, E.C., 2023. Ecosystem Status Report 2023: Eastern Bering Sea, Stock Assessment and Fishery Evaluation Report, North Pacific Fishery Management Council, 1007 West Third, Suite 400, Anchorage, Alaska 99501.
Whitworth, K., T. Vicente, A. Magel, K. Howard, V. von Biela, M. Williams, and P. Chambers. 2023. Factors Affecting 2023 Yukon & Kuskokwim Chum Salmon Runs and Subsistence Harvests. In: Siddon, E.C., 2023. Ecosystem Status Report 2023: Eastern Bering Sea, Stock Assessment and Fishery Evaluation Report, North Pacific Fishery Management Council, 1007 West Third, Suite 400, Anchorage, Alaska 99501.
Yasumiishi, E. 2023. Temperature Change Index and the Recruitment of Bering Sea Pollock. In: Siddon, E.C., 2023. Ecosystem Status Report 2023: Eastern Bering Sea, Stock Assessment and Fishery Evaluation Report, North Pacific Fishery Management Council, 1007 West Third, Suite 400, Anchorage, Alaska 99501.
Yasumiishi, E, L. Eisner, and D. Kimmel. 2023b. Large Copepod Abundance (Sample-Based and Modeled) as an Indicator of Pollock Recruitment to Age-3 in the Southeastern Bering Sea. In: Siddon, E.C., 2023. Ecosystem Status Report 2023: Eastern Bering Sea, Stock Assessment and Fishery Evaluation Report, North Pacific Fishery Management Council, 1007 West Third, Suite 400, Anchorage, Alaska 99501.
Yasumiishi, E., A. Andrews, J. Murphy, A. Dimond, and E. Farley. 2023c. Trends in the Biomass of Jellyfish in the South- and Northeastern Bering Sea During Late-Summer Surface Trawl Surveys, 2004–2023. In: Siddon, E.C., 2023. Ecosystem Status Report 2023: Eastern Bering Sea, Stock Assessment and Fishery Evaluation Report, North Pacific Fishery Management Council, 1007 West Third, Suite 400, Anchorage, Alaska 99501.
Appendix on model-based methods on bottom-trawl survey biomass trends
Overview
These applications of VAST were configured to model NMFS/AFSC bottom trawl survey (BTS) data and for acoustic backscatter data (next section). For the BTS, the station-specific CPUEs (kg per hectare) for pollock were compiled from 1982-2023. Further details can be found at the GitHub repo mainpage, wiki, and glossary. The R help files, e.g., ?make_data for explanation of data inputs, or ?make_settings for explanation of settings.
The software versions of dependent programs used to generate VAST estimates were:
- R (4.3.0)
- MKL libraries via Microsoft R Open (4.0.2)
- INLA (21.11.22)
- Matrix (1.4-0)
- TMB (1.9.6)
- TMBhelper (1.4.0)
- VAST (3.10.1)
- FishStatsUtils (2.12.1)
For these model-based index time series, we used the same VAST model run settings.
Spatio-temporal treatment of survey data on pollock density
For EBS pollock we used data on biomass per unit area from all grid cells and corner stations in the 83-112 bottom trawl survey of the EBS, 1982-2023, including exploratory northern extension samples in 2001, 2005, and 2006, as well as 83-112 samples available in the NBS in 1982, 1985, 1988, 1991, 2010, and 2017-2023 (except 2020). NBS samples collected prior to 2010 and in 2018 did not follow the 20 nautical mile sampling grid used in 2010, 2017, 2019, 2021, and 20232019, 2021–2023 surveys. Assimilating these data therefore required extrapolating into unsampled areas. As before, we included a a spatially varying covariate of the cold-pool extent (Thorson 2019, (O’Leary et al. 2020). All environmental data used as covariates were computed within the R package coldpool (Rohan et al., 2022).
We used a Poisson-link delta-model (Thorson 2018) involving two linear predictors and a gamma distribution to model positive catch rates. We extrapolated population density to the entire EBS and NBS in each year, using extrapolation grids that are available within [FishStatsUtils] (https://github.com/James-Thorson-NOAA/FishStatsUtils). These extrapolation grids were defined using 3705 m (2 nmi) × 3705 m (2 nmi) cells; this results in 36,690 extrapolation-grid cells for the eastern Bering Sea and 15,079 in the northern Bering Sea. We used bilinear interpolation to interpolate densities from 750 “knots” to these extrapolation grid cells; knots were approximately evenly distributed over space, in proportion to the dimensions of the extrapolation grid. We estimated geometric anisotropy (how spatial autocorrelation declines with differing rates over distance in some cardinal directions than others), and included a spatial and spatio-temporal term for both linear predictors. To facilitate interpolation of density between unsampled years, we specified that the spatio-temporal fields were structured over time as an AR(1) process (where the magnitude of autocorrelation was estimated as a fixed effect for each linear predictor). However, we did not include any temporal correlation for intercepts, which we treated as fixed effects for each linear predictor and year. Finally, we used epsilon bias-correction to correct for retransformation bias (Thorson and Kristensen 2016).
We checked model fits for evidence of non-convergence by confirming that (1) the derivative of the marginal likelihood with respect to each fixed effect was sufficiently small (less than ~0.001) and (2) that the Hessian matrix was positive definite. We then checked for evidence of model fit by computing Dunn-Smyth randomized quantile residuals (Dunn and Smyth 1996) and visualizing these using a quantile-quantile plot within the DHARMa R package. We also evaluated the distribution of these residuals over space in each year, and inspected them for evidence of residual spatio-temporal patterns.
Spatio-temporal treatment of survey age composition data
For model-based estimation of age compositions in the Bering Sea, we fitted observations of numerical abundance-at-age at each sampling location. This was made possible by applying a year-specific, region-specific (EBS and NBS) age-length key to records of numerical abundance and length-composition. We computed these estimates in VAST, assuming a Poisson-link delta-model (Thorson 2018) involving two linear predictors, and a gamma distribution to model positive catch rates. We did not include any density covariates in estimation of age composition for consistency with models used in the previous assessment, and due to computational limitations. We used the same extrapolation grid as implemented for abundance indices, but here we modeled spatial and spatiotemporal fields with a mesh with coarser spatial resolution than the index model, here using 50 “knots”. This reduction in the spatial resolution of the model, relative to that used abundance indices, was necessary due to the increased computational load of fitting multiple age categories and using epsilon bias-correction. We implemented the same diagnostics to check convergence and model fit as those used for abundance indices.
Densities and biomass estimates
Relative densities over time suggests that the biomass of pollock can reflect abundances in the NBS even in years where samples are unavailable (all years except 2010, 2017–2019 and 2021–2023; (). Index values and error terms (based on diagonal of covariance matrix over time) are shown in Figure 70.
Additional references
Dunn, K.P., and Smyth, G.K. 1996. Randomized quantile residuals. Journal of Computational and Graphical Statistics 5, 1-10.
Hartig, F. 2021. DHARMa: Residual Diagnostics for Hierarchical (Multi-Level / Mixed) Regression Models. R package version 0.4.0. http://florianhartig.github.io/DHARMa/
O’Leary, C.A., Thorson, J.T., Ianelli, J.N. and Kotwicki, S., 2020. Adapting to climate‐driven distribution shifts using model‐based indices and age composition from multiple surveys in the walleye pollock (Gadus chalcogrammus) stock assessment. Fisheries Oceanography, 29(6), pp.541-557.
Rohan, S.K., Barnett L.A.K., and Charriere, N. 2022. Evaluating approaches to estimating mean temperatures and cold pool area from AFSC bottom trawl surveys of the eastern Bering Sea. U.S. Dep. Commer., NOAA Tech. Mem. NMFS-AFSC-456, 42 p. https://doi.org/10.25923/1wwh-q418
Thorson, J.T., 2019. Measuring the impact of oceanographic indices on species distribution shifts: The spatially varying effect of cold‐pool extent in the eastern Bering Sea. Limnology and Oceanography, 64(6), pp.2632-2645.
Thorson, J.T., and Kristensen, K., 2016. Implementing a generic method for bias correction in statistical models using random effects, with spatial and population dynamics examples. Fisheries Research, 175, pp.66-74.

